Method for identifying bighead cod based on multispectral imaging system

Abstract

The present invention discloses a method for identifying bighead cod based on a multispectral imaging system, and relates to the technical field of fish identification. The present invention discloses a method for identifying bighead cod, and the identification method comprises the following steps: (1) obtaining spectral data information of a fish sample; (2) bringing the spectral data information into a trained QDA model to obtain a database; (3) bringing a fish sample of an unknown variety into the database obtained in step (1) for comparison to obtain a specific variety. The method for identifying bighead cod constructed by the present invention is simple to operate, has a short operation time, and can identify the authenticity of bighead cod without complex instruments and tedious comparison processes.

Description

Technical Field

[0001] The invention relates to the technical field of fish identification, and in particular to a method for identifying bighead cod based on a multispectral imaging system. Background Art

[0002] Gadus macrocephalus, also known as Pacific cod, belongs to the order Gabiformes, family Gadidae, genus Gadus. Gadus is mainly distributed in the North Pacific. It has an elongated, slightly flattened body, and a tail that tapers backwards. It is 25-40cm long, with a large head and mouth. The upper jaw is slightly longer than the lower jaw. There is a barbel on the neck, which is equal to or slightly longer than the eye diameter. The head, back and sides of the body are gray-brown, the belly is gray-white, the pectoral vein is light yellow, and the other fins are gray.

[0003] There are many methods for identifying fish species at present, which mainly include traditional morphological methods, protein analysis methods and molecular biological analysis methods according to their technical principles. The morphological identification method is the most traditional identification method, which is mainly based on morphological standards, such as appearance, color, smell and taste and other sensory indicators for identification. It has the advantages of being simple and intuitive, low cost and strong practicality. However, when it is used to identify fish samples with very similar characteristics, the identification method becomes difficult, which limits the use of this method.

[0004] The prior art discloses a real-time fluorescence PCR detection method for bighead cod ingredients, specifically, comparing the mitochondrial 16S rRNA gene sequences of 16 species of Gadformes fish and their easily confused species, such as bighead cod, Atlantic cod, and haddock, selecting primers or probes with strong specificity, and establishing a real-time fluorescence PCR detection method for cod ingredients. The absolute sensitivity and relative sensitivity of this method for detecting cod products are relatively strong, providing a strong guarantee for the trade of cod (Research on Real-time Fluorescence PCR Detection Technology of Cod and "Oily Fish" Ingredients, Xin Hongmei, Dalian University of Technology, 2023).

[0005] Literature: A method for detecting the components of codfish on the market was disclosed in the study of three molecular identification techniques for common codfish. Antarctic toothfish and snake mackerel, which are easily confused with cod, were selected as negative controls. The 16SrDNA gene was used to design primers to distinguish between codfish and non-codfish. The mitochondrial Cytb gene was used to design species-specific primers for Atlantic cod, Pacific cod, haddock, Antarctic toothfish and snake mackerel, respectively. A qPCR system was established to perform species identification analysis. The test results showed that the established qPCR system had good specificity and repeatability, and the 16S rDNAqPCR system could quickly and effectively distinguish between codfish and non-codfish species (Study of three molecular identification techniques for common codfish, Wang Yi, Zhejiang Gongshang University, 2020).

[0006] In the prior art, there are studies that use DNA microbarcoding combined with high-resolution melting curves (HRM) to identify cod species, and the main identified species include Atlantic cod, Pacific cod, pollock and blue cod. The study mainly conducted computer analysis on two barcode regions, namely cytochrome c oxidase subunit I (COI) and cytochrome b (cytb), to determine the genetic differences between the four species, and used them to develop a real-time PCR method combined with HRM analysis. However, this method can only identify cod species, and the detection method is relatively complicated.

[0007] At present, how to establish an identification method that is applicable to bighead cod and has simple operation methods to identify the authenticity of bighead cod is one of the most urgent needs in the market. Summary of the invention

[0008] The invention aims to provide a method for identifying bighead cod, which is simple and can identify bighead cod more accurately.

[0009] Terminology explanation:

[0010] The term "multispectral data" refers to the data collected within a specific spectral wavelength range, which is divided into 10 to 100 equal parts, which is called multispectral data.

[0011] The term "image segmentation" refers to the technology and process of dividing an image into several specific regions with unique properties and proposing objects of interest. It is a key step in image processing. Image segmentation is the process of dividing a digital image into non-overlapping regions. The image segmentation process is a labeling process, that is, pixels belonging to the same region are assigned the same number.

[0012] The term "multispectral imaging technology" is an emerging technology developed based on imaging and spectroscopy. It is an analytical tool that, based on the simultaneous labeling of multiple fluorescent signals, is excited by monochromatic light, so that multiple fluorescent signals are mixed together, and the required wavelength of light can be filtered and collected. Then, the system will unmix the collected multiple mixed lights, and after signal output and display, the different components or locations of biological samples marked with different colors can be intuitively observed. The multispectral imaging system can obtain high-resolution spectra of each pixel in each image.

[0013] The term "database" refers to a collection of data that is stored in a certain way and can be shared by multiple users. It is independent of the application program. The database has the following characteristics: it has as little duplication as possible, serves multiple applications of a specific organization in the best way, its data structure is independent of the application program that uses it, and the addition, deletion, modification and query of data are managed and controlled by unified software.

[0014] The term "spectral reflectance", also known as spectral reflectance factor or spectral reflectance, refers to the ratio of the spectral radiant flux of a wavelength reflected from an object in a direction defined by a specified solid angle under specific lighting conditions to the spectral radiant flux of a wavelength reflected from a completely diffuse reflecting surface under the same conditions.

[0015] To achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows:

[0016] In one aspect, the present invention provides a method for identifying bighead cod, comprising the following steps:

[0017] (1) Obtaining spectral data information of fish samples;

[0018] (2) Bringing the spectral data information into the trained QDA model to obtain a database;

[0019] (3) Bringing the fish meat sample of unknown species into the database obtained in step (1) for comparison to obtain the specific species.

[0020] Preferably, the wavelength range of the spectral data information collection described in step (1) is 19 bands from visible light to near infrared light, namely 365nm, 405nm, 430nm, 450nm, 470nm, 490nm, 515nm, 540nm, 570nm, 590nm, 630nm, 645nm, 660nm, 690nm, 780nm, 850nm, 880nm, 940nm, and 970nm.

[0021] Specifically, the extraction of spectral data information of the fish sample in step (1) includes: extracting the fish sample, photographing the fish sample using a multispectral imager to obtain a multispectral image of the fish sample, and extracting spectral data of the fish sample using a multispectral imaging system.

[0022] More specifically, the time for taking photos of the fish sample using the multispectral imager is 5-20 seconds;

[0023] To be more specific, the photo taking time is 5s-10s.

[0024] More specifically, the method of extracting the spectral data of fish meat using a multispectral imaging system comprises the following steps:

[0025] (1) Open VideometerLab software and import the multispectral image of the fish sample;

[0026] (2) Remove image background, perform image segmentation or perform layer annotation;

[0027] (3) Use VideometerLab software to extract spectral information of the selected area.

[0028] Preferably, the image segmentation in step (2) is performed using a fixed pixel size of 52*52 pixels.

[0029] In some implementations of the present invention, the pixel size of fixed pixels may be adjusted according to actual needs.

[0030] Preferably, the operation of step (3) in the VideometerLab software is to select Image Tools→Measurement→Statistics in sequence to obtain the spectral information of the sample.

[0031] Specifically, the spectral information is the average spectral reflectance value of the selected area.

[0032] Further preferably, the training of the QDA model includes the following steps: bringing the spectral data of the bighead cod into the QDA model, using the creation tool of the model, importing the training data through the model's own functions, loading the model and building the network, performing training, and obtaining training results.

[0033] Still more preferably, the model creation tool is RStudio.

[0034] More specifically, the QDA model is loaded into the model through the MASS package in R, and then the predict method is called for prediction.

[0035] Preferably, the training of the QDA model includes the step of splitting the spectral data.

[0036] Further preferably, the spectral data is specifically split into 70% of the spectral data as a training set and 30% of the spectral data as a test set.

[0037] Preferably, the QDA training code:

[0038] fit_qda_clsm<-qda(form_clsm,data=traindata,method="t").

[0039] Specifically, the data ranges of each spectrum of the bighead cod are: 365nm is 20.78-26.18, 405nm is 19.37-25.34, 430nm is 20.37-26.10, 450nm is 25.88-31.11, 470nm is 26.54-31.62, 490nm is 26.36-31.35, 515nm is 25.31-30.18, 540nm is 25.05-29.82, and 570nm is 23.39-28.1 5.590nm is 23.73-28.63, 630nm is 24.16-29.25, 645nm is 24.29-29.46, 660nm is 24.33-29.50, 690nm is 24.27-29.53, 780nm is 24.12-29.56, 850nm is 31.30-36.21, 880nm is 34.18-38.86, 940nm is 40.19-44.37, and 970nm is 40.86-45.03.

[0040] Specifically, the fish includes frozen fish or fresh fish.

[0041] Preferably, the sliced ​​meat is selected from the meat near the spine of the fish.

[0042] Specifically, the main purpose of sampling near the spine is to obtain larger and more uniform pieces of fish meat as possible.

[0043] Specifically, the size of the fish meat sample is (0.5-3cm)*(0.5-3cm)*(0.5-3cm).

[0044] Preferably, the training of the QDA model also includes a step of testing the training results, and the testing method is to use a confusion matrix to calculate the accuracy (Accuracy), recall rate (Macro-R), precision rate (Macro-P) and harmonic mean (Macro-F1) of the model.

[0045] The calculation formula is as follows:

[0046] Accuracy:

[0047] Accuracy:

[0048] Recall:

[0049] Where: True Positive (TP): The positive sample is successfully predicted as positive.

[0050] True Negative (TN): The negative sample is successfully predicted as negative.

[0051] False Positive (FP): Negative samples are mistakenly predicted as positive.

[0052] False Negative (FN): The positive sample is mistakenly predicted as negative.

[0053] F-number:

[0054] Macro calculation formula:

[0055]

[0056]

[0057]

[0058] The calculation result of the accuracy is the average value of each accuracy.

[0059] Preferably, the specific variety comparison in step (3) is to bring the sample data to be tested into the database model, obtain the predicted variety and probability value, and compare whether the class result and label in predict match. If they match, the prediction is correct, and if they do not match, the prediction is wrong.

[0060] The beneficial effects of the present invention are:

[0061] The identification method of the bighead cod constructed by the present invention is simple to operate, has a short operation time, and can identify the authenticity of the bighead cod without complicated equipment and tedious comparison process. The present invention provides a new method for identifying the bighead cod. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is the average spectrum and error data of bighead cod.

[0063] Figure 2 This is the confusion matrix diagram of the QDA model.

[0064] Figure 3 This is a graph showing the prediction results of using the QDA model to predict unknown fish meat samples. DETAILED DESCRIPTION

[0065] In order to make the technical means, creative features, purpose and efficacy of the present invention easy to understand, the present invention is further explained in conjunction with specific embodiments below, but the following embodiments are only preferred embodiments of the present invention, not all. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work all belong to the protection scope of the present invention. In the following embodiments, unless otherwise specified, the operating methods used are conventional operating methods, the equipment used are conventional equipment, and the equipment materials used in each embodiment are the same.

[0066] Example 1: Sample collection and spectral information extraction

[0067] The collection of samples includes the following steps:

[0068] (1) Select a cod to be processed, cut a fish meat sample of 1 cm*2 cm*1 cm near the spine of the cod, place the fish meat sample in a 50 mL centrifuge tube, and store the centrifuge tube at -80°C.

[0069] (2) Using a multispectrometer (the model of the multispectrometer is VideometerLab4), the fish meat samples in step (1) are scanned respectively, and the analysis time is 5s-10s to obtain a multispectral image of the cod meat sample.

[0070] (3) After scanning, the sample image is segmented using a fixed pixel size of 52*52 pixels.

[0071] Spectral information is extracted from the multispectral image of the cod meat sample, and the spectral information extraction method comprises the following steps:

[0072] The spectral information is extracted using the multi-spectral imager supporting software VideometerLab for reading. The process is as follows:

[0073] (1) Open VideometerLab software and import the multispectral image of the fish sample;

[0074] (2) Remove image background, perform image segmentation or perform layer annotation;

[0075] (3) VideometerLab software was used to extract spectral information of the selected area. The wavelength range of spectral information collection was 19 bands from visible light to near-infrared light, namely 365nm, 405nm, 430nm, 450nm, 470nm, 490nm, 515nm, 540nm, 570nm, 590nm, 630nm, 645nm, 660nm, 690nm, 780nm, 850nm, 880nm, 940nm and 970nm.

[0076] Step (3) is to select Image Tools→Measurement→Statistics in the VideometerLab software to obtain the spectral information of the sample. The mean and error of the average spectral reflectance value of the bighead cod under different spectral conditions are shown in Figure 2. Figure 1 The specific values ​​are shown in Table 1.

[0077] Table 1.

[0078] sp 1 2 3 4 5 6 7 8 Band365 27.85 22.45 23.65 25.68 22.51 23.79 25.62 24.63 Band405 25.72 21.36 21.99 23.75 21.71 23.03 24.9 24.52 Band430 26.77 22.24 22.83 24.8 22.58 23.76 25.57 25.2 Band450 33.07 26.73 28.27 31.23 26.64 28.69 30.45 28.96 Band470 33.87 27.26 28.93 32.01 26.93 29.04 30.73 29.07 Band490 33.74 27.11 28.78 31.86 26.6 28.65 30.3 28.65 Band515 32.61 26.25 27.78 30.7 25.44 27.31 28.94 27.41 Band540 32.16 26.01 27.58 30.37 25.24 27.06 28.6 27.12 Band570 30.51 24.51 25.85 28.73 23.57 25.03 26.72 25.24 Band590 31.02 24.92 26.27 29.31 23.93 25.27 27.07 25.47 Band630 31.67 25.49 26.76 29.99 24.47 25.62 27.56 25.86 Band645 31.89 25.71 26.94 30.21 24.66 25.7 27.71 25.95 Band660 31.91 25.77 26.92 30.26 24.7 25.64 27.74 25.96 Band690 31.86 25.84 26.84 30.24 24.82 25.57 27.68 25.86 Band780 31.58 26.05 26.69 30.11 25.09 25.23 27.53 25.79 Band850 38.05 32.88 33.52 36.87 32.78 32.39 34.67 33.23 Band880 40.55 35.64 36.33 39.56 35.93 35.27 37.46 36.23 Band940 45.59 41.68 42.06 44.89 42.77 41.49 43.42 42.8 Band970 45.94 42.52 42.7 45.31 44.01 42.34 44.19 43.92

[0079] Example 2 Model Training

[0080] (1) QDA model training

[0081] The spectral data in Example 1 was imported into the QDA model. The creation tool of the model, RStudio, was used to import the training data through the functions provided by the model. The model was loaded and the network was constructed. Training was performed to obtain training results.

[0082] The model training process is as follows:

[0083] ① Load spectral data;

[0084] ② Spectral data splitting, 70% of the spectral data is the training set, 30% of the spectral data is the test set, seed = 34;

[0085] ③Load the QDA model;

[0086] ④ Use the confusion matrix to reflect the results after training.

[0087] In step ②, Seed: the seed value set for random sampling.

[0088] The training code for QDA is:

[0089] fit_qda_clsm<-qda(form_clsm,data=traindata,method="t").

[0090] The model training results are as follows:

[0091] The confusion matrix of the QDA model is as follows Figure 2 shown.

[0092] The overall accuracy, recall, precision, and harmonic mean of the database model trained by the QDA model are shown in Table 2.

[0093] Table 2.

[0094] Model QDA Accuracy 0.9906 Recall (Macro-R) 0.9855 Precision (Macro-P) 0.9898 Harmonic mean (Macro-F1) 0.9875

[0095] As can be seen from Table 2 above, the accuracy of the QDA model is 99.06%, the recall rate is 0.9855, the precision rate is 0.9898, and the harmonic mean is 0.9875.

[0096] Example 3 Identification of unknown samples

[0097] The identification process for unknown fish samples during testing includes the following steps:

[0098] (1) Load the data to be predicted and the data labels;

[0099] (2) Load the model and call the predict function (R language tool);

[0100] The data label in step (1) is the label of the unknown sample, and its main purpose is to distinguish different samples.

[0101] The experimental results are verified by comparing whether the class result and label in predict match. If they match, the prediction is correct, if not, the prediction is wrong.

[0102] The prediction accuracy of the three groups of data is shown in Table 3 below:

[0103] Table 3.

[0104]

[0105]

[0106] From the above Table 3, it can be seen that when the number of data is 195, the prediction accuracy of the QDA model is 98.74%. When the number of data is 1905, the prediction accuracy of the QDA model is 98.74%. When the number of data is 2045, the prediction accuracy of the QDA model is 98.63%.

[0107] The prediction process of unknown fish samples includes the following steps:

[0108] (1) Load the data to be predicted;

[0109] (2) Load the model and use the predict function to predict;

[0110] (3) Obtain the predicted varieties and probability values.

[0111] Experimental results:

[0112] Load the unknown fish sample into the trained model and get the prediction result Figure 3No. 15 represents cod, which means that the spectral data information of the unknown fish sample falls into the spectral database of cod constructed above, and the unknown sample can be determined to be cod.

[0113] The above codes are shown in Table 4: Codes.

[0114] Table 4.

[0115]

[0116]

[0117] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying bighead cod, characterized in that: The following steps are involved: (1) Obtaining spectral data information of fish samples; (2) Bring the spectral data information into the trained QDA model to obtain a database; (3) Bringing the fish meat sample of unknown species into the database obtained in step (1) for comparison to obtain the specific species; The training of the QDA model includes the following steps: bringing the spectral data of the bighead cod into the QDA model, using the creation tool of the model, importing the training data through the functions provided by the model, loading the model and building the network, training, and obtaining the training results; The wavelength range of the spectral data information collection described in step (1) is 19 bands from visible light to near infrared light, namely 365nm, 405nm, 430nm, 450nm, 470nm, 490nm, 515nm, 540nm, 570nm, 590nm, 630nm, 645nm, 660nm, 690nm, 780nm, 850nm, 880nm, 940nm, and 970nm; The data ranges of each spectrum of the bighead cod are: 365nm is 20.78-26.18, 405nm is 19.37-25.34, 430nm is 20.37-26.10, 450nm is 25.88-31.11, 470nm is 26.54-31.62, 490nm is 26.36-31.35, 515nm is 25.31-30.18, 540nm is 25.05-29.82, 570nm is 23.39-28.15, 590nm is 23.73-28.63, 630nm is 24.16-29.25, 645nm is 24.29-29.46, 660nm is 24.33-29.50, 690nm is 24.27-29.53, 780nm is 24.12-29.56, 850nm is 31.30-36.21, 880nm is 34.18-38.86, 940nm is 40.19-44.37, 970nm is 40.86-45.03; The spectral data are average spectral reflectance values.

2. The identification method according to claim 1, characterized in that: The model is created using RStudio.

3. The identification method according to claim 1, characterized in that: The QDA model is loaded into the model by the MASS package in R, and then the predict method is called for prediction.

4. The identification method according to claim 1, characterized in that: The training code of the QDA model is: fit_qda_clsm <- qda(form_clsm, data = traindata, method = "t").

5. The identification method according to claim 1, characterized in that: The training of the QDA model includes a step of testing the training results, and the testing method is to use a confusion matrix to calculate the accuracy, recall, precision and harmonic mean of the model for testing.

6. The identification method according to claim 1, characterized in that: The comparison is to bring the unknown fish sample data into the database model, obtain the predicted species and probability value, and compare whether the class result and label in predict match. If they match, the prediction is correct, otherwise, it is an error.

Images