TWI749437B - Method and apparatus for identifying a seafood sample and method for determining a freshness of a seafood sample - Google Patents

Abstract

A method and apparatus for field spectroscopic characterization of seafood is disclosed. A portable NIR spectrometer is connected to an analyzer configured for performing a multivariate analysis of reflection spectra to determine qualitatively the true identities or quantitatively the freshness of seafood samples.

Description

Method and device for identifying a seafood sample and method for judging the freshness of a seafood sample

The present invention relates to material properties and identification, and specifically relates to the spectroscopic properties of seafood.

One of the largest investigations recently published a report aimed at revealing information about seafood fraud (one-third of seafood purchased in restaurants and grocery stores in cities across the United States were mislabeled). Oceana (a non-profit international advocacy group) conducted a two-year study from 2010 to 2012, and collected more than 1,200 samples from 674 retail stores in 21 states in the United States (February 2013, K. Warner, W "Oceana Study Reveals Seafood Fraud Nationwide" report by Timme, B. Lowell and M. Hirshfield). Perform DNA testing on fish samples to correctly identify fish species and find mislabeling. Similar conclusions can be obtained from one of the previous US Congressional Research Services report on combating fraud and deception in seafood marketing (Congressional Research Services Report No. 7-5700, www.crs.gov, RL-34124 (2010)). It is illegal to replace a more expensive fish line with a low-cost species. This is because monetary gains motivate criminals, leading to negative economic, health, and environmental consequences. Consumers and honest seafood suppliers have been tricked into paying higher prices for lower-cost undesirable alternatives. One of the most common alternative and more expensive fish is the red snapper, which is usually replaced by tilapia. In addition, some fish substitutes cause health hazards. For example, the Oceana study above has determined that more than 90% of the fish advertised as white tuna are actually escolar, which contains one of the toxins known to cause gastrointestinal problems, snake mackerel Species. Finally, some alternative fish may be an overfished or threatened species. One fish is Atlantic cod, which was found to be replaced with Pacific cod in the same study. The supply chain "from ship to disk" is complex and unadjusted, making such illegal activities difficult to track. Fighting fish fraud requires traceability of fish supply across the entire supply chain and increased inspections. DNA testing is time-consuming and can only be performed on a sampling basis. DNA testing requires taking fish samples to a laboratory and waiting for the results (a procedure can take several days). Wong's US Patent No. 5,539,207 discloses a method for identifying human or animal tissues by Fourier Transform Infrared (FT-IR) spectroscopy. The mid-infrared spectrum of one of the tissues under discussion is measured and compared with an infrared spectrum library of known tissues to find a closest match. A visual comparison or a pattern recognition algorithm can be used to match the infrared spectrum. In this way, various tissues (and even normal tissues or malignant (e.g. cancer) tissues) can be identified. Disadvantageously, Wong's method is difficult to use for the purpose of seafood identification in field conditions. An FT-IR spectrometer is a complex and huge optical device. The core module of the FT-IR spectrometer (a scanning Michelson interferometer) uses a large, precise and movable optical mirror to perform a wavelength scan. To stabilize the mirror, use a heavy optical bench. Due to many precise optical and mechanical components, an FT-IR spectrometer requires laboratory conditions and requires frequent recalibration and realignment by trained personnel. The use of an FT-IR spectrometer is specified by the fact that the fundamental vibration frequency of infrared fingerprints is present in the 2.5 micron to 5 micron region of the electromagnetic spectrum. These vibration frequency bands have high resolution and high absorption levels, and exhibit strong absorption with narrow spectral bands. Monro's US Patent No. 7,750,299 discloses a system for active biometric spectroscopy, in which a frequency-tunable millimeter wave radio transmitter irradiates a DNA membrane of a specific organism, and detects transmission and scattering by the DNA membrane Of radio waves. Monro teaches that the radio wave scattering spectra of different DNA membranes are different. Therefore, the transmitted or scattered radio wave spectrum can detect different DNA membranes, which can be associated with different fish species. In this way, the species of a fish sample can be identified. Disadvantageously, Monro's method cannot be applied to fish samples themselves, because the signal from non-DNA tissue will overwhelm the DNA signal. Because of this, the DNA of the fish sample needs to be extracted and formed into a membrane. Sample preparation is time-consuming and can only be performed in laboratory conditions. US Patent No. 7,728,296 to Cole et al. discloses a device and method for detecting explosive materials using terahertz (THz) radiation. THz radiation occupies the frequency band between infrared and millimeter radio waves. Many explosive materials have one of the only spectral symbols in the THz frequency domain, thus giving high sensitivity and non-invasive remote detection of explosives. Disadvantageously, THz radiation sources are bulky and expensive, limiting their current use for critical applications such as airport checkpoints. The methods and devices of the prior art are not suitable for the purpose of identifying seafood species in the field conditions. There is a need for a method and system to enable a Food and Drug Administration (FDA) official to perform a rapid on-site seafood species identification and characterization to assist the official in deciding whether to take an enforcement action. Individuals (such as restaurant chefs, sushi chain patrons and fish market customers) will also benefit from the possibility of quickly verifying purchased seafood species.

One object of the present invention is to provide a method and device for on-site spectroscopic characteristics of seafood. From a technical point of view, it is better to perform wavelength band spectroscopy for easy growth, wavelength separation, and electromagnetic radiation detection. A near-infrared (NIR) frequency band (for example, between 0.7 micrometers and 2.5 micrometers) satisfies this condition. Broadband LEDs and even small incandescent sources can be used to generate NIR light in this wavelength band. Various spectrally selective elements (such as thin film interference filters) can be used for wavelength separation. The photodiode array can be used to detect NIR light. Although it is convenient to work in the NIR part of the spectrum, the previous technology has mostly focused on longer and less technology-friendly wavelength bands. This is because the main vibration frequency of the characteristic molecular bonds of most organic compounds corresponds to a frequency greater than 2.5 microns (2500 The long wavelength of nanometers necessitates the use of heavier and bulky equipment to generate wavelength dispersion and to detect these longer-wavelength electromagnetic radiation. The inventor has realized that multiple vibration frequencies or so-called overtones are indeed in the technically convenient NIR band, and therefore, although this information is hidden due to multiple frequencies of a relatively low amplitude and overtones, biometric identification information Appears in the NIR spectrum. When it is not easy to obtain or visually recognize spectroscopic information from a spectrum, advanced data processing and feature or pattern extraction and modeling techniques (such as principal component analysis (PCA), classification and analog soft independent modeling (SIMCA), partial least square discrimination) Analysis (PLS-DA) and Support Vector Machine (SVM)) can be used to extract the required information. Therefore, multi-variable type identification and data return realize the use of a lightweight and compact NIR spectrometer to identify and characterize seafood. According to the present invention, there is provided a method for identifying a seafood on site, which includes: (a) Provide a portable NIR spectrometer; (b) Use the NIR spectrometer of step (a) to obtain the reflectance spectrum of one of the seafood samples; (c) Perform a multi-variable type identification analysis of the reflectance spectra of the seafood samples obtained in step (b) to compare the reflectance spectra with a spectral library of known identification spectra corresponding to different species of seafood, and Determine the matching spectrum with one of the most similar spectral patterns; and (d) Identify the seafood sample based on the matching spectrum with the most similar spectrum pattern determined in step (c). These pattern recognition algorithms can also generate a confidence measure or a probability estimate of a possibility of the recognition result. According to the present invention, there is further provided a method for judging the freshness of a seafood sample on site, which includes: (a) Provide a portable NIR spectrometer; (b) Use the NIR spectrometer of step (a) to obtain the reflectance spectrum of one of the seafood samples; (c) Perform a multivariable type identification analysis of the reflectance spectrum of the seafood sample obtained in step (b) to compare the reflectance spectrum with a spectrum library of known identity spectra corresponding to the freshness of the seafood sample, and It is determined that there is a matching spectrum of the most similar spectrum pattern, thereby providing a quantitative measurement of the freshness of the seafood sample. The reflectance spectrum can be obtained from multiple positions on the seafood sample to reduce the influence of the surface texture of the seafood sample. Multivariate regression analysis can include, for example, partial least squares (PLS) and support vector regression (SVR). According to the present invention, there is further provided a device for identifying a seafood sample on site, which includes: A portable NIR spectrometer, which is used to obtain the NIR reflectance spectrum of one of the seafood samples, and An analyzer that is operatively coupled to a spectrometer and is configured to perform a multivariable type identification analysis of the reflectance spectrum of a seafood sample by comparing the reflectance spectrum with the known identification spectra of seafood corresponding to different species A spectrum library is determined to have a matching spectrum with the most similar spectrum pattern, and the seafood sample is identified based on the matching spectrum with the most similar spectrum pattern. The portable NIR spectrometer can include a spectral laterally variable optical transmission filter coupled to a light detector array, resulting in a particularly compact and lightweight structure. A mobile communication device can be configured to communicate with the NIR spectrometer and perform multivariate analysis of the reflection spectrum obtained by the portable NIR spectrometer. In addition, at least some data analysis and spectral pattern model building activities can be performed at a remote server that communicates with the mobile device. According to another aspect of the present invention, there is further provided a permanent storage medium which is installed in a mobile communication device and has a known identity spectrum library encoded on it.

下文，僅簡單討論表1之數值方法，此係由於該等方法其等本身在技術中已知。該等方法之各者具有其之優點。在單純貝氏(Naïve Bayes)方法中，假定全部特彼此獨立，且結果可容易解釋。CART方法亦易於理解及解釋；然而，自數值資料集產生之樹可為複雜的，且該方法趨於具有過度擬合問題。TreeBagger分析及隨機森林分析方法通常基於非常好的結果，且該方法之「訓練」步驟相對較快。LIBLINEAR方法在區分海產物種及條件中非常有效率。使用線性核心之SVM方法(包含用於定量分析之支援向量分類(SVC)及用於定量分析之支援向量迴路歸(SVR))導致超過93%之預測成功率。在LDA方法中，假定全部類別具有相同協方差矩陣且經正常分佈，且判別函數始終為線性的。在QDA方法中，該等類別無需具有相同協方差矩陣，但仍假定正常分佈。偏最小平方(PLS)係一統計方法，其帶有一些關於主分量迴歸；而不是找出反應與獨立變數之間之最小方差之超平面，其藉由將所預測之變數及可觀察變數投影至一新空間而找出一線性迴歸模型。偏最小平方判別分析(PLS-DA)係當Y類目時使用之一變體。PLS-DA方法導致85%至87%之適度預測率。 結果展示：NaiveBayes、TreeBagger、SVM-linear、LDA、QDA、PLS-DA及SIMCA可出於使NIR反射光譜與海產樣本相關之目的而用於多變數分析中。所獲得之光譜之第一導數及第二導數亦可用於取代、或除了光譜之預處理之外，作為多變數分析之一輸入資料串(data strings)。 用於實施結合文中所揭示之態樣而描述之各種繪示性邏輯、邏輯區塊、模組及電路之硬體可使用一通用處理器、一數位信號處理器(DSP)、一專用積體電路(ASIC)、一場可程式化閘陣列(FPGA)或其他可程式化邏輯器件、離散閘或電晶體邏輯、離散硬體組件或經設計以執行文中所描述之功能之其之任何組合來實施或執行。一通用處理器可為一微處理器，但在替代實施例中，處理器可為任何習知處理器、控制器、微控制器或狀態機。一處理器亦可作為計算器件之一組合(例如一DSP及一微處理器、複數個微處理器、結合一DSP核心之一或多個微處理器或任何其他此組態之一組合)而實施。替代地，一些步驟或方法可由特定於一給定功能之電路執行。 已出於說明及描述之目的而呈現本發明之一或多個實施例之以上描述。不意欲詳盡或將本發明限於所揭示之精確形式。諸多修改及變動可能係鑑於以上教示。由隨附於此之申請專利範圍來限制本發明之範疇，而非意欲由此詳細描述限制本發明之範疇。Although the teaching is described in conjunction with various embodiments and examples, it is not intended that the teaching is limited to these embodiments. On the contrary, this teaching includes various alternatives and equivalents, as those familiar with the art will understand. 1, a device 10 for identifying a seafood sample 11 on site includes a portable NIR spectrometer 12 for obtaining a diffuse NIR reflectance spectrum 13 (signal power P relative to wavelength λ) of the seafood sample 11. An analyzer 14 is operatively coupled to the spectrometer 12 via a cable 15. The analyzer 14 is configured to perform a multivariate analysis of the reflection spectrum 13 of the seafood sample 11 to determine at least one characteristic parameter corresponding to the reflection spectrum 13. The analyzer 14 is configured to compare at least one parameter with a threshold value corresponding to a species of the seafood sample 11 for determining the species of the seafood sample 11. These species can be displayed on a display 16 of the analyzer 14. At least one parameter may include two or more parameters. The two parameters can be represented graphically as a point on an XY diagram called a Coomas diagram. One of the points on the Coomans diagram represents the seafood species using reflectance spectrum 13. We will further consider multivariate regression/pattern identification analysis and Coomas diagrams below. First consider the structure of the NIR spectrometer 12. 2, the NIR spectrometer 12 includes a main body 23, an incandescent lamp 24 for illuminating a seafood sample 11, a tapered light pipe (TLP) 25 for guiding the diffuse reflected light 36, and a tapered light pipe (TLP) 25 for dividing the reflected light 36 into individual parts. A laterally variable filter (LVF) 31 of wavelengths and a photodetector array 37 for detecting optical power levels of individual wavelengths. The photodetector array 37 is formed in a CMOS processing chip 37A and is coupled to the LVF 31 using an optically transparent adhesive 38. An electronic board 37B is provided to support and control the CMOS processing chip 37A. An optional button 21 is provided to start spectrum collection. The photodetector array 37 is vertically aligned with one of the longitudinal axis LA of the TLP 25. In operation, the incandescent lamp 24 illuminates the seafood sample 11. The TLP 25 collects the diffusely reflected light 36 and directs it toward the LVF 31. The LVF 31 divides the diffuse reflection light 36 into individual wavelengths, and the light detector array 31 detects the individual wavelengths. The measurement cycle can be started by pressing the button 21 or by an external command from the analyzer 14. The compact NIR spectrometer 12 is realized by the structure of its photodetector assembly 29. Referring to FIG. 3A, the photodetector sub-assembly 29 is shown in the XZ plane. In FIG. 3A, the photodetector subassembly 29 is turned over 180 degrees as indicated by the direction of the z-axis on the right side of FIGS. 2 and 3A. In the preferred embodiment shown in FIG. 3A, the optically transparent adhesive 38 directly couples the photodetector array 37 to the LVF 31. The optically transparent adhesive 38 must be: in fact non-conductive or dielectric; by using the induced pressure or destructive force to the detector array 37 to achieve good adhesion strength and be mechanically neutral; optically compatible to transmit the desired spectral content ; Remove the reflection of air on the glass interface; and have a thermal expansion coefficient property to minimize the pressure to the detector pixel 52 during the thermal cycle. An opaque epoxy resin 22 encapsulates the LVF 31, facilitating the removal of stray light and protecting the LVF 31 from humidity. An optional glass window 39 is placed on top of the LVF 31 for additional environmental protection. Referring to FIG. 3B and FIG. 3C, the operation of the LVF 31 is shown. The LVF 31 is shown in the YZ plane where the wavelengths are dispersed. The LVF 31 includes a wedge-shaped spacer 32 sandwiched between the wedge-shaped dichroic mirrors 33 to form a Fabry-Perot interferometer, with a laterally variable interval between the dichroic mirrors 33 . The wedge shape of the optical transmission filter 31 makes its transmission wavelength laterally variable. For example, arrows 34A, 34B, and 34C are used to point to a transmission spectrum 35 (FIG. 3C) shown below the variable optical transmission filter 31, respectively. The individual transmission peaks are shown in 35A, 35B and 35C. In operation, the multi-color light 36 reflected from the seafood sample 11 is irradiated on the variable optical filter 31, and the variable optical filter 31 divides the multi-color light 36 into individual spectral components indicated by arrows 34A to 34C. The wavelength range of the NIR spectrometer 12 is preferably between 700 nm and 2500 nm, and more preferably between 950 nm and 1950 nm. The use of LVF 31 and TLP 25 allows the size of the NIR spectrometer 12 to be reduced considerably. The NIR spectrometer 12 does not have any moving parts for wavelength scanning. The small weight of the NIR spectrometer 12, which is generally less than 100 grams, allows the NIR spectrometer 12 to be placed directly on the seafood sample 11. The small weight and size also make the NIR spectrometer 12 (for example) easy to transport in a pocket of a marine inspector. The size of the NIR spectrometer 12 is shown in FIG. 3D. The NIR spectrometer 12 can be easily hand-held, and the button 21 is conveniently positioned for thumb operation. Many variants of the NIR spectrometer are certainly possible. For example, the incandescent bulb 24 can be replaced with a broadband light-emitting diode or LED. The TLP 25 can be replaced with another optical element, such as a fiber optic panel or a holographic beam shaper. The LVF 31 can be replaced with another suitable wavelength selection element (such as a small diffraction grating, an array of dichroic mirrors, a MEMS device, etc.). Referring to FIG. 4A and further referring to FIG. 1, a method 40 for identifying a seafood sample 11 on site includes a step 41 of providing the portable NIR spectrometer 12 described above. In a step 42, the NIR spectrometer 12 is used to obtain the reflectance spectrum 13 of the seafood sample 11. In a step 43, a multivariable pattern identification analysis of the reflection spectrum 13 of the seafood sample 11 is performed to determine by comparing the reflection spectrum 13 with a spectrum library of known identification spectra corresponding to different species of seafood A matching spectrum with one of the most similar spectral patterns. Finally, in a step 44, the seafood sample 11 is identified based on the matching spectrum with the most similar spectral pattern determined in the previous step 43. In the text, the term "match spectrum" does not automatically indicate an exact match. Alternatively, the “matched spectrum” indicates that the spectrum library carries one of the most similar spectral patterns as an identity spectrum as compared with the measured reflectance spectrum 13. Therefore, "match" is not the exact match of the available matches, but only the closest match. The proximity of the match can be calculated based on the specific match evaluation method used. Perform multi-variable type identification analysis 43 to extract seafood species information from the reflectance spectrum 13. Due to the many overtones of the vibrational frequencies of the characteristic molecular bonds, the reflection spectrum 13 can be very complex, making individual peaks of the spectrum unable to be visually recognized. According to the present invention, a multivariate type analysis 43 (also referred to as “chemometric analysis”) is performed to identify or identify the species of the seafood sample 11. The measuring step 42 preferably includes: performing repeated spectral measurements at different positions on the seafood sample 11; and averaging the repeated measurements to reduce the dependence of the obtained reflection spectrum on a texture of the seafood sample 11 . The extended multiplicative scattering correction (EMSC) of the reflection spectrum 13 can be used to reduce the dependence of the measured reflection spectrum 13 on the scattering properties of the seafood sample 11. Other known statistical methods can also be used to preprocess the reflectance spectrum 13, for example, the standard normal variation (SNV) of the reflectance spectrum 13 can be calculated before proceeding to the step 43 of identifying and analyzing the multivariate pattern. The Savitzky-Golay filter of the reflection spectrum 13 can be performed and the first and/or a second derivative of the reflection spectrum 13 can be calculated to be considered in the multivariable pattern identification analysis step 43, and the spectrum in the reflection spectrum 13 can be considered Characteristic slope and/or recurve. Other statistical methods (such as sample-by-sample normalization of the reflectance spectrum 13 and/or channel-by-channel automatic scaling) can be used to facilitate the multi-variable pattern identification and analysis step 43, and to provide more stable results. The multi-variable pattern identification analysis 43 is usually performed in two stages. For example, referring to FIG. 4B and further referring to FIG. 1, a PCA step 45 is first performed to define a calibration model of each seafood type to be identified. The PCA step 45 can be completed in advance before the seafood sample 11 is measured in a calibration phase of the device 10. In a second step 46, the similarity between the collected reflectance spectrum 13 and the calibration model of different seafood species is analyzed. In the illustrated embodiment, Soft Independent Modeling of Classification Analogs (SIMCA) is used. Due to SIMCA step 46, two parameters are determined. These two parameters are plotted in an XY graph (Coomans graph), and different areas of it correspond to different seafood species. In some cases, only one parameter is needed, and this parameter can be compared with a threshold determined in step 45 of the PCA to identify the seafood sample 11. Other multi-variable pattern identification and analysis methods can be applied. Consider examples of these methods in the "Experimental Verification" section below. In view of the proliferation of computerized mobile communication devices (such as smart phones), it is advantageous to use a mobile communication device to perform the multi-variable pattern identification and analysis step 43 (FIG. 4A and FIG. 4B). Referring to FIG. 5A and further referring to FIGS. 1 and 4A, a device 50A for identifying a seafood sample 11 on site is similar to the device 10 of FIG. 1. One difference in the device 50A of FIG. 5A is that a mobile communication device 54 is configured to perform the multivariate analysis step 43 and the identification step 44 of the method 40 of FIG. 4A. For this purpose, the mobile communication device 54 may include a permanent storage medium 58, which encodes a spectral library and/or computer commands corresponding to the known identification spectra of different species of seafood on it for executing multivariable patterns. Identification/data reduction analysis step 43. The mobile communication device 54 can be coupled to the NIR spectrometer 12 via a wireless link 59 (such as Bluetooth™) or via a wire (such as USB communication) for transmitting the obtained reflection spectrum 13 to the mobile communication device 54. Turning now to FIG. 5B and further referring to FIGS. 4A and 5A, a device 50B for identifying a seafood sample on site is similar to the device 50A of FIG. 5A. The apparatus 50B of FIG. 5B includes a remote server 57 that communicates with the mobile communication device 54 via an RF communication link 56 to a base station 55 (which is connected to the Internet 52). In operation, the self-mobile device 54 transmits the reflection spectrum 13 to the remote servo 57, and performs the multivariable pattern identification analysis at the remote server 57 (ie, step 43 of the method 40 in FIG. 4A). The result of the multivariate analysis step 43 (FIG. 4A) is sent back to the mobile device 54 (FIG. 5B) for display to a user (not shown). The identification step 44 (FIG. 4A) can be performed by the mobile device 54 or by the remote server 57 (FIG. 5B). Using the computing power of a remote server eliminates the need for resources on mobile communication devices, and therefore speeds up the overall process of seafood identification. Experimental verification Many experiments are performed to verify similar appearance, but a combination of NIR spectroscopy and multivariate regression (chemometrics) analysis can be used to identify fish species with different prices. Referring to Figures 6 to 8, three groups of different fish species are used. The first group contains: a whole red mullet 60A and a whole mullet 60B (Figure 6), both skin and meat (the meat is not shown). The second group includes: winter cod skin 71A; cod skin 71B; winter cod meat 72A; and cod meat 72B. The third group includes: young salmon skin 81A; European trout skin 81B; young salmon meat 82A; and European trout 82B. As can be seen from the photos from Figure 6 to Figure 8, even for a seafood professional (such as a wholesaler or a chef, not to mention general public customers), visually distinguishing the whole fish and fish meat will be quite challenging. In FIGS. 6 to 8, the "A" group includes more expensive species 60A, 71A, 72A, 81A, and 82A, and the "B" group includes less expensive species 60B, 71B, 72B, 81B, and 82B. Therefore, the use of "B" species instead of "A" species can provide a substantial economic benefit. Turning to FIG. 9, one of the devices 90 used in the experimental verification of the present invention includes a miniature NIR™ 1700 spectrometer 92 manufactured by JDS Uniphase, Milpitas, California, USA. The miniature NIR spectrometer 92 operates in a wavelength range of 950 nm to 1650 nm. The miniature NIR spectrometer 92 is a low-cost, extremely compact and portable spectrometer, which weighs 60 grams and is less than 50 mm in diameter. The spectrometer 92 operates in a diffuse reflection and is constructed similarly to the spectrometer 12 of FIG. 3B, and includes a light source (not shown) for illuminating the seafood sample 11, a dispersion element 31, a photodetector array 37, and electronics (Not shown) (all of them are contained in a small portable package that can be placed directly on a seafood sample 91). The spectrometer 92 is connected by a cable 95 to a laptop 94 running Unscrambler™ multivariate analysis software (provided by CAMO AS of Oslo, Norway (version 9.6)). For each spectrum measurement, 50 scans with an integration time of 5 milliseconds have been accumulated, resulting in a total measurement time of one 0.25 second per reflection spectrum measurement. Referring now to FIGS. 10A and 10B, flowcharts 100A and 100B show the steps of spectrum acquisition and PCA model building performed for fish samples 60A and 60B; 71A and 71B; 72A and 72B; 81A and 81B; and 82A and 82B, respectively. In steps 101A and 101B, three different individual species are provided to the fish samples 60A and 60B; 71A and 71B; 72A and 72B; 81A and 81B; 82A and 82B in each of FIGS. 6 to 8 respectively. For mullet 60A and 60B; winter cod/cod 71A and 71B; 72A and 72B; and young salmon/European trout 81A and 81B; 82A and 82B pairs, collect skin reflectance spectra in steps 102A and 102B, respectively; and in step 103A And 103B collected meat reflectance spectra. Obtaining a whole of ten NIR reflection spectra at different positions on each of the three pieces, resulting in the fish samples 60A; 60B; 71A; 71B; 72A; 72B; 81A;; 81B; 82A and Figures 6 to 8 Thirty measurements of 82B. One of the standard methods of extended multiplicative scatter correction is used to correct the spectrum for scatter. Therefore, all thirty spectra have been obtained for each fish skin type 60A and 60B; 71A and 71B; 81A and 81B in steps 104A and 104B, respectively. All thirty spectra have been obtained for each fish type 72A and 72B; 82A and 82B in steps 105A and 105B, respectively. The spectra have been averaged into five groups for each of the three samples of each type in steps 106A, 107A; and 106B, 107B, resulting in two average spectra for each sample and six average spectra for each sample type , Including skin and meat. Averaging is completed to reduce the dependence of the obtained reflection spectrum on the texture of the respective seafood samples 60A, 60B, 71A, 71B, 72A, 72B, 81A, 81B, 82A, and 82B. Next, the PCA model has been established in steps 108A and 108B for the respective "A" and "B" samples. Perform a SIMCA analysis to identify the type of each fish sample. The results are presented according to the Coomas diagram of each fish type. Red mullet/mullet pairing Refer to Figure 11 and further to Figure 6, the reflectance spectra of red mullet 60A and mullet 60B are shown as a reflection signal (arbitrary unit) pair between 10900 cm ^-1 and 6000 cm ^-1 The dependence of the wave number in the range (cm ^{-1 ).} In 111, twelve traces including the six spectra of red mullet skin and the six spectra of mullet skin are displayed. At 112, twelve traces including the six spectra of red mullet flesh and the six spectra of mullet flesh are displayed. It can be seen that the spectrum 111 of red mullet and mullet skin are very similar to each other, and the spectrum 112 of red mullet and mullet meat are also very similar to each other, so the spectrum of red mullet and mullet cannot be visually distinguished from the spectrum of mullet (for skin and meat Both). Turning to FIG. 12 and further referring to FIGS. 10A and 10B, the results of the PCA analysis steps 108A, 108B (FIG. 10B) are presented. In FIG. 12, the red mullet skin score 121A is sufficient to separate from the mullet skin score 121B to allow easy identification, but a clear separation is not achieved between the red mullet flesh score 122A and the mullet flesh score 122B. Now referring to Figures 13A and 13B, the results of the SIMCA analysis of the red mullet/mullet pairing are presented in the form of a 5% significant Coomas graph. Figure 13A shows the result of red mullet sample identification. The gray circle 131A represents the calibration red mullet sample (skin and meat) used to obtain the identity spectrum of the red mullet; the white filled circle 131B represents the calibration mullet sample (skin and meat) used to obtain the identity spectrum of the mullet; and The filled (black) circle 132 represents the test sample. All four black circles correspond to a sample of red mullet skin and a sample of red mullet flesh, each represented by two average spectra. Figure 13B shows the result of identification of mullet samples. The filled (black) circle 133 represents two test samples. All eight black circles 133 correspond to two mullet skin samples and two mullet flesh samples, each represented by two average spectra as explained above. Only one of the two parameters "distance to red mullet" and "distance to mullet" can be used by comparing the parameter with a threshold value. For example, if the "distance to mullet" is used, the threshold is about 0.01. If the "distance to red mullet" is used, the threshold is about 0.0008. It can be seen from Figures 13A and 13B that the red mullet (both skin and meat) can be easily identified. Therefore, removing the skin of a fish sample will not allow a potential criminal to conceal an illegal act of replacing red mullet with mullet. For winter cod/cod pairing, refer to Figure 14 and further refer to Figure 7. The reflectance spectra of winter cod skin 71A, winter cod meat 72A, cod skin 71B, and cod meat 72B (Figure 7) are shown as a reflection signal (arbitrary unit) pair at 10900 The dependence of the wave number in the range between cm ^-1 and 6000 cm ^{-1 (cm -1 ).} In 141, twelve traces including the six spectra of winter cod skin and the six spectra of cod skin are displayed. In 142, twelve traces including the six spectra of the respective winter cod meat and the six spectra of the cod meat are displayed. It can be seen that the spectra 141 of winter cod and cod skin are very similar to each other, and the spectra of winter cod and cod meat are also very similar to each other, so the spectrum of winter cod is visually indistinguishable from that of cod (for both skin and meat samples). Turning to FIG. 15 and further referring to FIGS. 10A and 10B, the results of the PCA analysis steps 108A, 108B (FIG. 10B) are presented. In Figure 15, the winter cod skin score 151A appears alternately interspersed with the cod skin score 151B, and the winter cod meat score 152A appears alternately interspersed with the cod meat score 152B, so a clear distinction cannot be made at this stage. Referring now to Figures 16A and 16B, the results of the SIMCA analysis of winter cod/cod pairing are presented in the form of a 5% significant Coomas graph. Figure 16A shows the result of identification of cod samples. The gray circle 161A represents the calibration winter cod sample (skin and meat) used to obtain the identity spectrum of the winter cod; the white filling circle 161B represents the calibration cod sample (skin and meat) used to obtain the identity spectrum of the cod; and the filling (black) Circle 162 represents the test sample. All eight black circles correspond to two cod skin samples and two cod meat samples, each represented by two average spectra as explained above. Figure 16B shows the results of winter cod sample identification. The filled (black) circle 163 represents a test sample. All four black circles 163 correspond to a winter cod skin sample and a winter cod meat sample, each represented by two average spectra. It can be seen from Figure 16A and Figure 16B that winter cod (both skin and meat) can be easily identified and distinguished from cod. Juvenile salmon / Salmon pairing with reference to Figure 17 and further reference to Figure 8, parr skin 81A, juvenile salmon meat 82A, European trout skin 81B and European trout meat reflection spectrum by display 82B of the reflection signal (in arbitrary units) of the 10900 cm ^- The dependence of the wave number in the range of ¹ to 6000 cm ^{-1 (cm -1 ).} In 171, twelve traces including six spectra of young salmon skin and six spectra of European trout skin are displayed. In 172, twelve traces including the six spectra of each young salmon meat and the six spectra of European trout meat are displayed. It can be seen that the skin spectra 171 of young salmon and European trout are very similar to each other, and the flesh spectrum 172 of young salmon and European trout are also very similar to each other, so the spectrum of young salmon cannot be visually distinguished from the spectra of European trout (for both skin and meat samples) By). Turning to FIG. 18 and further referring to FIGS. 10A and 10B, the results of the PCA analysis steps 108A, 108B (FIG. 10B) are presented. In Figure 18, the young salmon skin score 181A appears to be alternately interspersed with the European trout skin score 181B, and the young salmon meat score 182A appears to be alternately interspersed with the European trout score 182B, so a clear distinction cannot be completed at this stage. Now referring to Figures 19A and 19B, the results of the SIMCA analysis of the young salmon/European trout pairing are presented in the form of a 5% significant Coomas chart. Figure 19A shows the results of the identification of a trout sample. The gray circle 191A represents the calibration salmon sample (skin and meat) used to obtain the identification spectrum of the young salmon; the white filling circle 191B represents the calibration salmon sample (skin and meat) used to obtain the identification spectrum of the salmon; and the filling ( The black circle 192 represents the test sample. All eight black circles correspond to two European trout skin samples and two European trout meat samples, each represented by two average spectra. Figure 19B shows the results of the identification of the young salmon sample. The filled (black) circle 193 represents two test samples. All four black circles 193 correspond to two young salmon skin samples and two young salmon meat samples, each represented by two average spectra. It can be seen from Figure 19A and Figure 19B that young salmon (both skin and meat) can be easily identified and distinguished from European trout. The freshness of mullet (Meerbarbe) fillets has been performed on a numerical study of the reflectance spectra of mullet fillets, in which various known multivariate analysis methods are used to distinguish between the freshness conditions of mullet fillets (both skin and skinless meat). The following Table 1 summarizes the success prediction rate of the alternative matching method of mullet and red mullet executed on a typical desktop computer. The spectrum is automatically scaled before being sent to the multi-variable type classification. The time to perform prediction based on the model is usually in the range of milliseconds. When an on-site model update is required, the time to build the model can be used as a determining factor. In the field, the shorter the point-of-use application, the speed of measurement, and the speed of obtaining results, the more important. In addition, the accuracy of the results is very important. From Table 1, it can be seen that methods such as SVM (using a linear core) provide the best accuracy in the shortest time. Table 1

In the following, only the numerical methods of Table 1 are briefly discussed, because these methods themselves are known in the art. Each of these methods has its advantages. In the Naïve Bayes method, all characteristics are assumed to be independent of each other, and the results can be easily interpreted. The CART method is also easy to understand and explain; however, the tree generated from a numerical data set can be complex, and the method tends to have an overfitting problem. TreeBagger analysis and random forest analysis methods are usually based on very good results, and the "training" step of the method is relatively fast. The LIBLINEAR method is very efficient in distinguishing seafood species and conditions. The use of linear core SVM methods (including support vector classification (SVC) for quantitative analysis and support vector loop regression (SVR) for quantitative analysis) resulted in a prediction success rate of over 93%. In the LDA method, it is assumed that all categories have the same covariance matrix and are normally distributed, and the discriminant function is always linear. In the QDA method, these categories do not need to have the same covariance matrix, but normal distribution is still assumed. Partial least squares (PLS) is a statistical method with some regressions on principal components; instead of finding the hyperplane of the minimum variance between the response and the independent variables, it uses the projection of the predicted variable and the observable variable To a new space and find a linear regression model. Partial Least Squares Discriminant Analysis (PLS-DA) is a variant used when the Y category is used. The PLS-DA method resulted in a moderate prediction rate of 85% to 87%. Results show: NaiveBayes, TreeBagger, SVM-linear, LDA, QDA, PLS-DA and SIMCA can be used in multivariate analysis for the purpose of correlating NIR reflectance spectra with seafood samples. The first derivative and the second derivative of the obtained spectrum can also be used in place of, or in addition to preprocessing of the spectrum, as one of the input data strings for multivariate analysis. The hardware used to implement various graphical logics, logic blocks, modules, and circuits described in combination with the aspects disclosed in the text can use a general-purpose processor, a digital signal processor (DSP), and a dedicated integrated body Circuit (ASIC), a programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination of them designed to perform the functions described in the text are implemented Or execute. A general-purpose processor may be a microprocessor, but in alternative embodiments, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor can also be used as a combination of computing devices (for example, a DSP and a microprocessor, a plurality of microprocessors, a combination of one or more microprocessors of a DSP core, or any other combination of this configuration). Implement. Alternatively, some steps or methods can be performed by a circuit specific to a given function. The foregoing description of one or more embodiments of the present invention has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and changes may be based on the above teachings. The scope of the present invention is limited by the scope of the patent application attached hereto, and it is not intended to limit the scope of the present invention by this detailed description.

10: device 11: Seafood samples 12: Portable near infrared light (NIR) spectrometer 13: Diffuse near infrared light (NIR) reflectance spectrum 14: Analyzer 15: Cable 16: display 21: Button 22: Opaque epoxy resin 23: main body 24: Incandescent lamp/Incandescent bulb 25: Tapered Light Pipe (TLP) 29: Light detection sub-assembly 31: Horizontal variable filter (LVF)/dispersion element 32: Wedge spacer 33: Wedge dichroic beam splitter 34A: Arrow 34B: Arrow 34C: Arrow 35: Transmission spectrum 35A: Individual transmission peak 35B: Individual transmission peak 35C: Individual transmission peak 36: Diffuse light/multicolor light 37: Light detector array 37A: Complementary Metal Oxide Semiconductor (CMOS) processing wafer 37B: Electronic board 38: Optically transparent adhesive 39: use glass windows 40: Method 41: Steps 42: Measurement steps 43: Multi-variable type identification analysis/data reduction analysis steps 44: identification steps 45: Principal Component Analysis (PCA) step 46: The second step/soft independent modeling (SIMCA) step of classification analogy 50A: device 50B: Device 52: Detector pixel/Internet 54: mobile communication device 55: base station 56: Radio Frequency (RF) Communication Link 57: remote server 58: Permanent storage media 59: wireless link 60A: Red mullet/fish sample 60B: Mullet/fish sample 71A: Winter Cod Skin 71B: Cod skin 72A: Winter Cod Meat 72B: Cod meat 81A: young salmon skin 81B: European trout skin 82A: young salmon 82B: European trout 90: device 91: Seafood samples 92: Mini NIR™ 1700 Spectrometer 94: laptop 95: Cable 100A: Flow chart 100B: Flow chart 101A: Step 101B: Step 102A: Step 102B: Step 103A: Step 103B: Steps 104A: Step 104B: Step 105A: Step 105B: Step 106A: Step 106B: Step 107A: Step 107B: Step 108A: Principal Component Analysis (PCA) analysis steps 108B: Principal Component Analysis (PCA) analysis steps 111: Spectrum 112: Spectrum 121A: Red mullet skin score 121B: Mullet skin score 122A: Red mullet score 122B: Mullet score 131A: Calibration red mullet sample (skin and meat) used to obtain the identity spectrum of red mullet 131B: Calibration mullet sample (skin and meat) used to obtain the identity spectrum of mullet 132: test sample 133: Mullet skin sample and mullet meat sample 141: Winter cod skin / cod skin spectrum 142: Winter cod meat / cod meat spectrum 151A: Winter cod skin score 151B: Cod skin score 152A: Winter Cod Meat Score 152B: Cod meat score 161A: Calibration winter cod sample (skin and meat) 161B: Calibration cod sample (skin and meat) 162: test sample 163: test sample 171: Young salmon skin/European trout skin spectrum 172: Young salmon meat/European trout meat spectrum 181A: Young salmon skin score 181B: European trout skin score 182A: Young salmon score 182B: European trout score 191A: Calibration young salmon sample (skin and meat) 191B: Calibration sample of European trout (skin and meat) 192: Test sample 193: test sample λ: wavelength LA: Longitudinal axis P: signal power XZ: plane YZ: plane z: axis

Illustrative embodiments will now be described in conjunction with the drawings, in which: Figure 1 is a schematic three-dimensional diagram of a device used to identify a seafood sample on site according to the present invention, in which a NIR reflectance spectrum measured by the device is overprinted in the figure; Figure 2 is a side cross-sectional view of a portable handheld NIR spectrometer of the device of Figure 1; Fig. 3A is a side cross-sectional view of a light detecting sub-assembly of the portable NIR spectrometer of Fig. 2; FIG. 3B is a side cross-sectional view of a wavelength dispersing element used in the light detecting sub-assembly of FIG. 3A; Figure 3C is a transmission spectrum of one of the wavelength separation elements of Figure 3B; Figure 3D is a three-dimensional view of the portable handheld NIR spectrometer of Figure 2; 4A is a flowchart of a method for identifying a seafood sample on site according to the present invention; 4B is a flowchart of an exemplary multivariate analysis of NIR spectra according to the present invention; FIG. 5A is a schematic diagram of an embodiment of the device of the present invention, in which a portable device wirelessly communicating with the NIR spectrometer is used to analyze the NIR spectrum obtained by the NIR spectrometer; 5B is a schematic diagram of another embodiment of the device of the present invention, in which a portable device is used to relay the measured NIR spectrum to a remote server for performing multivariate analysis; Figures 6 to 8 are examples of red mullet/mullet pairing (Figure 6), winter cod/cod pairing (skin and meat, Figure 7) and young salmon/European trout pairing to be used in the experimental verification of the present invention. Color photos of the seafood pairings distinguishing between skin and meat, Fig. 8); Figure 9 is a color photo of a prototype of a device for measuring the NIR spectrum of a salmon sample; Figures 10A and 10B are flowcharts of data collection and analysis of higher and lower quality seafood used for experimental verification, respectively; Figure 11, Figure 14 and Figure 17 are the measured diffuse reflectance spectra of red mullet/mullet pairing, winter cod/cod pairing and young salmon/European trout pairing respectively; Figures 12, 15 and 18 are the three-dimensional score graphs of the measured principal component analysis (PCA) model of red mullet/mullet pairing, winter cod/cod pairing and young salmon/Euro trout pairing; and Figure 13A, Figure 13B; Figure 16A, Figure 16B, Figure 19A, Figure 19B are red mullet/mullet pairing; winter cod/cod pairing; and young salmon/European trout pairing classification analogy soft independent modeling (SIMCA ) Analysis of the Coomans diagram.

10: device

11: Seafood samples

12: Portable near infrared light (NIR) spectrometer

13: Diffuse near infrared light (NIR) reflectance spectrum

14: Analyzer

15: Cable

16: display

Claims (20)

A method for providing information about a sample, the method comprising: receiving a reflection spectrum of the sample by a mobile device; providing the reflection spectrum to a remote server by the mobile device; using the mobile device and A result of a multi-variable type identification analysis is received from the remote server. The multi-variable type identification analysis is based on the reflection spectrum and a known identification spectrum corresponding to different seafood species by comparing a distance parameter with a The threshold value is executed with the threshold value associated with one of the different seafood species; the mobile device determines the information about the sample based on the result of receiving the multivariable type identification analysis; by The mobile device provides and is used to display the information about the sample. For example, the method of claim 1, wherein receiving the result of the multi-variable pattern identification analysis includes: receiving the multi-variable pattern identification analysis after the remote server recognizes the sample based on executing the multi-variable pattern identification analysis The result. For example, the method of claim 1, wherein determining the information about the sample includes: identifying the sample based on the multi-variable type identification analysis by the mobile device and after receiving the result of the multi-variable type identification analysis; And determine the information about the sample based on identifying the sample. The method of claim 1, wherein the mobile device receives the reflection spectrum from a portable spectrometer connected to the mobile device via a wireless link. The method of claim 1, wherein providing the reflection spectrum to the remote server includes: providing the reflection spectrum to the remote server via a communication link to a base station. Such as the method of claim 1, wherein the sample is a seafood sample. Such as the method of claim 1, wherein the information about the sample includes an identity of the sample, and a spectrum library in which the known identity spectrum is generated is based on the first plural number of at least one seafood sample collected for the plurality of samples Performing an operation on the first plurality of spectra to generate a second plurality of spectra corresponding to the known identity spectrum. A mobile device includes: one or more memories; and one or more processors, which are communicatively coupled with the one or more memories, and the one or more processors are configured to: from a spectrometer Receive a reflection spectrum of a sample; provide the reflection spectrum to a remote server; A result of a multi-variable type identification analysis is received from the remote server. The multi-variable type identification analysis is based on the reflection spectrum and a known identification spectrum corresponding to different seafood species by comparing a distance parameter with a Implementation of the threshold value, the threshold value being associated with one of the different seafood species; and providing information about the sample based on the result of the multivariate type identification analysis received for display. Such as the mobile device of claim 8, wherein when receiving the result of the multi-variable type identification analysis, the one or more processors are configured to: perform the multi-variable type identification analysis based on the remote server After identifying the sample, the result of the multivariable pattern identification analysis is received. Such as the mobile device of claim 8, wherein the one or more processors are further configured to: after receiving the result of the multi-variable type identification analysis, identify the sample based on the multi-variable type identification analysis. Such as the mobile device of claim 8, wherein the mobile device and the spectrometer are connected via a wireless link, and the spectrometer is a portable spectrometer. Such as the mobile device of claim 8, wherein when providing the reflection spectrum to the remote server, the one or more processors are configured to: The reflection spectrum is provided to the remote server via a base station. Such as the mobile device of claim 8, wherein the sample is a seafood sample. For example, the mobile device of claim 8, wherein the information about the sample includes an identity of the sample, and a spectrum library of known identity spectra generated therein is based on the first of at least one seafood sample collected for the plurality of samples A plurality of spectra and performing an operation on the first plurality of spectra to generate a second plurality of spectra corresponding to the known identity spectrum. A permanent computer-readable medium for storing instructions. The instructions include: one or more instructions, which when executed by one or more processors, cause the one or more processors to: receive a reflection from a sample Spectrum; provide the reflection spectrum to a remote server; receive a result from the remote server based on a multi-variable type identification analysis, the multi-variable type identification analysis is based on the reflection spectrum and corresponding to different seafood The known identity spectrum of the species is performed by comparing a distance parameter with a threshold value, the threshold value being associated with one of the different seafood species; and based on the result of receiving the multivariate type identification analysis Provide information about the sample for display. For example, the permanent computer-readable medium of claim 15, wherein the one or more instructions that cause the one or more processors to receive the result of the multivariable pattern identification analysis cause the one or more processors to: After the remote server recognizes the sample based on executing the multi-variable pattern identification analysis, it receives the result of the multi-variable pattern identification analysis. For example, the permanent computer-readable medium of claim 15, wherein when the one or more instructions are executed by the one or more processors, the one or more processors: After analyzing the result, the sample is identified based on the multi-variable pattern identification analysis. For example, the permanent computer readable medium of claim 15, wherein the generation of a spectrum library of known identity spectra is based on the first plurality of spectra collected for the plurality of samples of at least one seafood sample and the An operation is performed on each spectrum to generate a second plurality of spectra corresponding to the known identity spectrum. For example, the permanent computer-readable medium of claim 15, wherein the reflection spectrum is received by a portable spectrometer. For example, the permanent computer readable medium of claim 19, wherein the one or more processors are included in a mobile device, and the mobile device is connected to the portable spectrometer via a wired link.

Images