A method for identifying bivalve molluscs by combining molecular identification with three-dimensional vision

Abstract

The invention discloses a bivalve mollusc identification method combining molecular identification with three-dimensional vision. Firstly, a bivalve mollusc sample is dissected and divided into adductor muscle tissue and shell, DNA of the adductor muscle tissue is extracted, COI gene sequence is amplified and sequenced by PCR, and molecular tags of the bivalve mollusc sample are obtained; then, the shell of the shell is restored, photos of the shell of the shell at different angles are taken by a three-dimensional model shooting device, and the photos are processed to obtain a three-dimensional model of the bivalve mollusc sample; then, a model for classification and identification is obtained based on three-dimensional model data training, and finally, the three-dimensional model data of the shell sample to be tested is input into the model. The model can realize classification and identification of unknown shells containing only three-dimensional model data, and the classification accuracy rate is above 80%, which solves the relevant difficulties of identification of morphologically similar shells and non-invasive identification, and has practical application significance for improving the breeding and production efficiency of bivalve molluscs, as well as the identification of cryptic species in ecological detection.

Description

Technical field:

[0001] The invention belongs to the technical field of shellfish species classification and identification, and specifically relates to a bivalve shellfish identification method combining molecular identification with three-dimensional vision. Background technology:

[0002] Bivalve molluscs are one of the main species of marine aquaculture in my country and have important economic value in coastal areas. Therefore, the morphological identification of molluscs has become a key issue in breeding, ecological monitoring and protection. However, in the classification of mollusc breeding screening or cryptic species monitoring, the measurement of morphological indicators based on traditional methods still has certain limitations in classification. The main reasons are: ① It is difficult to fully describe bivalve molluscs as three-dimensional objects based on the morphological information contained in two-dimensional morphological indicators. ② Traditional analysis methods such as principal component analysis (PCA) dimensionality reduction will lose some information contained in the data, resulting in inaccurate results. ③ Molecular identification methods require invasive sampling of samples, which may cause sample death. Therefore, there is an urgent need for a non-invasive shellfish morphological measurement method that can more accurately classify and identify species.

[0003] At present, with the development of three-dimensional vision technology and the reduction of the difficulty of obtaining three-dimensional models, many fields are using 3D vision for various applications. Three-dimensional models have high application value in bivalve morphology measurement because they are closer to real appearance simulation. In addition, as one of the hottest technologies in the past two years, machine learning can complete more accurate and efficient species classification tasks. The advantages of machine learning in three-dimensional data processing are mainly as follows: ① 3D species classification involves multidimensional data such as three-dimensional traits, textures, and structures. Machine learning can effectively process such high-dimensional data, thereby better capturing the differences between species. ② Machine learning can automatically extract and select the most discriminative features in 3D data, which helps to overcome the subjectivity and uncertainty when selecting features manually. ③ The self-learning and adaptability of machine learning algorithms can continuously improve classification performance based on new data, which is very beneficial for handling changing species classification tasks. ④ Machine learning algorithms can still show good robustness when facing data of poor quality, which is very beneficial for processing 3D data of varying quality. Therefore, the species classification method based on three-dimensional models combined with machine learning has important practical significance.

[0004] Cytochrome C oxidase subunit I (COI) is located on the mitochondrial genome. It is the gene with the largest molecular weight and the most conserved functional domain among the 13 protein-coding genes in the mitochondrial genome. At the same time, the COI gene can distinguish species with distant divergence times due to its slow evolution rate of amino acids. The COI gene sequences of species of the same genus have sufficient variation to distinguish species with close genetic relationships. Therefore, the COI gene is widely used in molecular systematic studies at different taxonomic levels. In addition, the COI gene can be easily amplified by universal primers while ensuring sufficient variation, and the 5' end sequence of most animals can be obtained. There are few insertions or deletions in its sequence, so the COI gene is selected as a marker gene for DNA classification; at the same time, it has a very broad application prospect in species identification using PCR technology.

[0005] The patent with publication number CN105586404A discloses a method for identifying shellfish germplasm by combining morphology and molecular biology. The method uses a combination of molecular biology methods and morphological methods to identify shellfish. Shellfish samples with certain morphological characteristics are accurately identified through molecular biology methods. The morphological characteristics that can be used for species identification are summarized, and a morphological characteristic retrieval table is established. The morphological characteristic retrieval table is then used in combination with a machine learning model to identify the species of other samples, and molecular biology methods are used for verification. This can significantly improve the accuracy of shellfish germplasm resource identification and solve many difficulties in previous species identification. Summary of the invention:

[0006] The purpose of the present invention is to provide a non-invasive and accurate species classification method by combining bivalve molecular identification with a three-dimensional model and a machine learning algorithm.

[0007] In order to achieve the above-mentioned object, the present invention relates to a method for identifying bivalve molluscs by combining molecular identification with three-dimensional vision, which specifically comprises the following steps:

[0008] (1) dissecting a bivalve mollusc sample, dividing it into adductor muscle tissue and shell, extracting DNA from the adductor muscle tissue, amplifying the COI gene sequence by PCR and sequencing, and obtaining a molecular tag of the bivalve mollusc sample, wherein the molecular tag refers to the species to which the bivalve mollusc sample belongs;

[0009] (2) restoring the shell of the bivalve mollusc obtained in step (1), taking photos of the shell of the bivalve mollusc at different angles using a three-dimensional model shooting device, and processing the photos to obtain a three-dimensional model of the bivalve mollusc sample;

[0010] (3) The nine morphological indicators of the 3D model are extracted by the bounding box method, namely length L, width W, height H, length L / width W, height H / length L, width W / height H, ellipticity E, and sphericity and mean curvature H, the above morphological indices are integrated to obtain a morphological index data set;

[0011] (4) According to the molecular identification results, the species category is annotated after the corresponding morphological index data set to complete the construction of the machine learning model data set, and then the data set is divided into a training set and a test set. The training set is input into the KNN model and the random forest model RF for training, and the test set is input into the trained KNN model and the random forest model RF for testing. The model training results are evaluated using classification accuracy;

[0012] (5) obtaining a morphological index data set of the shellfish sample to be tested according to steps (2)-(3), and inputting the data set into the trained KNN model or random forest model RF, and the results show that the data set can be accurately classified;

[0013] or

[0014] (3') labeling the three-dimensional model file data of the shellfish sample in step (2) according to the species to which the shellfish sample belongs by previous molecular identification, and then dividing the data into a training set and a test set, inputting the training set into the PointNet++ model to train the model, and inputting the test set into the trained PointNet++ model to test it, and evaluating the model training results using classification accuracy;

[0015] (4') According to step (2), the three-dimensional model file data of the shellfish sample to be tested is obtained and input into the trained PointNet++ model. The results show that it can accurately classify it.

[0016] Step (1) is specifically as follows:

[0017] (101) The dissected adductor muscle tissue was placed in anhydrous ethanol solution for storage to prevent muscle tissue degradation;

[0018] (102) About 50 mg of adductor muscle tissue was taken and placed in a sterile centrifuge tube. 500 μL of lysis buffer was added to the centrifuge tube, followed by 5 μL of proteinase K. The tube was mixed by inversion and placed in a 40 °C water bath for digestion overnight.

[0019] (103) A mixture of phenol, chloroform, and isoamyl alcohol in a volume ratio of 25:24:1 was added to the centrifuge tube, which was then placed in a rotating incubator and mixed for 20 min;

[0020] (104) Place the centrifuge tube in a high-temperature refrigerated centrifuge and centrifuge at 12,000 rpm for 10 min at 4 °C;

[0021] (105) Transfer the supernatant after centrifugation in step (104) to a new centrifuge tube, add 600 ml of isopropanol, and precipitate at -20°C for more than 30 minutes;

[0022] (106) Centrifuge at 12000 rpm for 10 min. A white precipitate is observed at the bottom. Discard the supernatant and keep the white precipitate at the bottom.

[0023] (107) Add 75% ethanol solution to the white precipitate and centrifuge at 12,000 rpm for 10 min. Discard the supernatant and retain the precipitate. Repeat this process until a white to transparent precipitate is obtained, which is the DNA of the adductor muscle tissue.

[0024] (108) The precipitate obtained at the end of step (107) was air-dried, and about 50 μL of TE buffer was added to dissolve the DNA. The OD value and concentration value of the DNA solution were detected using a UV spectrophotometer, and the DNA solution was detected by electrophoresis. Figure 4 ;

[0025] (109) The COI target fragment of the bivalve sample was amplified using a PCR amplification instrument. The reaction system was 25 μL, including 12.5 μL PCR mix premix, 10 μL H2O, 1 μL of upstream and downstream primers, and 0.5 μL template DNA. The upstream and downstream primer sequences were:

[0026] LCO1490:5′-GGTCAACAAATCATAAAGATATTGG-3′;

[0027] HCO2198:5-TAAACTTCAGGGTGACCAAAAAATCA-3′;

[0028] (110) The COI gene fragment of the amplified mitochondrial DNA is sequenced, and the sequencing results are compared with the NCBI database to determine the taxonomic status of the shellfish to be identified and obtain a molecular tag, which refers to the species to which the bivalve shellfish sample belongs.

[0029] The three-dimensional model shooting device in step (2) includes a shooting device 1, a lighting system 2, a 360° stage 4 and a background plate 5. The bivalve mollusc sample 3 is placed on the 360° stage 4. The 360° stage 4 is used to drive the shooting device to rotate 360° at a constant speed during shooting. The shooting device 1 is used to take photos of one side, obliquely above and obliquely below the bivalve mollusc sample; the background plate 5 is used to ensure that the background of the photo is a pure black background to ensure image quality; the position of the lighting system 2 is used to provide the light source required for shooting to improve the brightness of the shooting.

[0030] Step (2) is specifically as follows:

[0031] (201) Restore the shellfish shell obtained in step (1) and place it on a stage, turn on the stage power switch, and keep the stage rotating 360°;

[0032] (202) Turn on and adjust the camera 1 to ensure that the lens is in focus, so that the camera 1 is at a position 45° above the shell of the shellfish, adjust the position of the background plate 5 to ensure that the background of the photograph is a pure black background, rotate the stage 360°, and collect photos of different positions of the shell of the shellfish;

[0033] (203) The photographing device 1 is made level with the shell of the shellfish, and the other steps are the same as (202), and the stage is rotated 360° to collect photos of different positions of the shell of the shellfish;

[0034] (204) The photographing device 1 is positioned 45° below the shell of the shellfish, and the other steps are the same as (202). The stage is rotated 360° to collect photos of different positions of the shell of the shellfish;

[0035] (205) Collect the images taken in steps (202)-(204), and use photogrammetry software (e.g., RealityCapture) to process and model them to obtain a three-dimensional model of the sample.

[0036] Step (3) is specifically as follows:

[0037] (301) Convert the 3D model file into a corresponding point cloud file. Each vertex in the point cloud file includes 3D coordinates, color and other information. Use the Open3D library in Python to read the 3D model file and calculate the size of its axis-aligned bounding box. Then use the size of the bounding box as the length L, width W and height H of the 3D model, and calculate the ratios between them: length / width, height / length and width / height;

[0038] (302) Calculate the ellipticity E of the three-dimensional model, Wherein, L is the length of the three-dimensional model, H is the height of the three-dimensional model, and W is the width of the three-dimensional model;

[0039] (303) Calculate the sphericity of a three-dimensional model Where: S represents the surface area of ​​the three-dimensional model, and V represents the volume of the three-dimensional model;

[0040] (304) Calculate the average curvature H of the three-dimensional model. Use the TriMesh library to calculate the average curvature by calculating the curvature at each vertex of the three-dimensional model, and then average and normalize to obtain the curvature characteristics of each three-dimensional model. The curvature of vertex i is Mean curvature Among them, θ i represents the angle between the normals of the adjacent faces of vertex V, A i Represents the area of ​​adjacent faces.

[0041] Compared with the prior art, the present invention has the following beneficial effects: (1) capturing the three-dimensional structural data of shellfish from multiple viewpoints is conducive to obtaining richer structural information of samples; (2) the KNN model, RF model and PointNet++ model constructed based on molecular identification and three-dimensional vision can accurately classify the species of bivalve samples that only collect three-dimensional model information but not COI molecular information, especially the classification and identification of cryptic species, and the classification accuracy rate is above 80%; (3) the present invention solves the difficulties related to the identification of morphologically similar shellfish and non-invasive identification, and has practical application significance for improving the breeding and production efficiency of bivalve shellfish, as well as the identification of cryptic species in ecological detection. Description of the drawings:

[0042] Figure 1 This is a flow chart of the bivalve mollusc identification method combining molecular identification with three-dimensional vision involved in Example 1.

[0043] Figure 2 This is a schematic diagram of the structure of the three-dimensional model shooting device involved in Example 1.

[0044] Figure 3 This is a schematic diagram of the three-dimensional model of bivalve molluscs obtained after being processed by the three-dimensional model shooting device involved in Example 1.

[0045] Figure 4 This is the result of DNA electrophoresis of the bivalve mollusc samples involved in Example 1.

[0046] Figure 5 These are photos of two bivalves of the northern and southern species involved in Example 1.

[0047] Figure 6 It is the accuracy of the KNN model trained with different proportions of training sets involved in Example 1.

[0048] Figure 7 The AUC values ​​of the KNN model and the RF model trained with different proportions of training sets involved in Example 1.

[0049] Figure 8 The importance of the three-dimensional morphological indicators involved in Example 1 in the RF model.

[0050] Fig. 9 It is the accuracy of the PointNet++ model trained with different proportions of training sets involved in Example 2. Specific implementation method:

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0052] Example 1

[0053] like Figure 1 As shown, the present embodiment relates to a method for identifying bivalve molluscs by combining molecular identification with three-dimensional vision, comprising the following steps:

[0054] (1) The bivalve samples were dissected and separated into adductor muscle tissue and shell. DNA from the adductor muscle tissue was extracted. The COI gene sequence was amplified by PCR and sequenced to obtain the molecular tags of the bivalve samples.

[0055] (101) The dissected adductor muscle tissue was placed in anhydrous ethanol solution for storage to prevent muscle tissue degradation;

[0056] (102) About 50 mg of adductor muscle tissue was taken and placed in a sterile centrifuge tube. 500 μL of lysis buffer was added to the centrifuge tube, followed by 5 μL of proteinase K. The tube was mixed by inversion and placed in a 40 °C water bath for digestion overnight.

[0057] (103) A mixture of phenol, chloroform, and isoamyl alcohol in a volume ratio of 25:24:1 was added to the centrifuge tube, which was then placed in a rotating incubator and mixed for 20 min;

[0058] (104) Place the centrifuge tube in a high-temperature refrigerated centrifuge and centrifuge at 12,000 rpm for 10 min at 4 °C;

[0059] (105) Transfer the supernatant after centrifugation in step (104) to a new centrifuge tube, add 600 ml of isopropanol, and precipitate at -20°C for more than 30 minutes;

[0060] (106) Centrifuge at 12000 rpm for 10 min. A white precipitate is observed at the bottom. Discard the supernatant and keep the white precipitate at the bottom.

[0061] (107) Add 75% ethanol solution to the white precipitate and centrifuge at 12,000 rpm for 10 min. Discard the supernatant and retain the precipitate. Repeat this process until a white to transparent precipitate is obtained, which is the DNA of the adductor muscle tissue.

[0062] (108) The precipitate obtained at the end of step (107) was air-dried, and about 50 μL of TE buffer was added to dissolve the DNA. The OD value and concentration value of the DNA solution were detected using a UV spectrophotometer, and the DNA solution was detected by electrophoresis. Figure 4 ;

[0063] (109) The COI target fragment of the bivalve sample was amplified using a PCR amplification instrument. The reaction system was 25 μL, including 12.5 μL PCR mix premix, 10 μL H2O, 1 μL of upstream and downstream primers, and 0.5 μL template DNA. The upstream and downstream primer sequences were:

[0064] LCO1490:5′-GGTCAACAAATCATAAAGATATTGG-3′;

[0065] HCO2198:5-TAAACTTCAGGGTGACCAAAAAATCA-3′;

[0066] (110) The COI gene fragment of the amplified mitochondrial DNA is sequenced, and the sequencing results are compared with the NCBI database to determine the taxonomic status of the shellfish to be identified and obtain a molecular tag, which refers to the species to which the bivalve shellfish sample belongs.

[0067] Specifically, this embodiment selects 224 bivalve samples (all samples belong to Bivalvia, genus V. mangle) from 10 populations of Huludao (HLD), Zhuanghe (ZH), Dalian (DL), Rushan (RS), Rongcheng (RC), Lianyungang (LYG), Yancheng (YC), Dongtai (DT), Zhoushan (ZS) and Xiamen (XM) with extremely similar external morphology, and each sample is processed in step (1). According to the COI structure, the above 10 populations are divided into northern species (V. atrata) and southern species (V. mangle), the northern species include Huludao (HLD), Zhuanghe (ZH), Dalian (DL), Rushan (RS), Rongcheng (RC), Lianyungang (LYG), and the southern species include Yancheng (YC), Dongtai (DT), Zhoushan (ZS) and Xiamen (XM), and the nucleotide difference between the two species is 100bp.

[0068] Figure 5 These are photos of bivalves of the northern species (V.atrata) and the southern species (V.mangle). The two species are very similar in appearance, and according to the COI results, they have large genetic differences and are cryptic species.

[0069] (2) The shell of the bivalve mollusc obtained in step (1) is restored, and photos of the shell of the bivalve mollusc at different angles are taken using a three-dimensional model shooting device, and the photos are processed to obtain a three-dimensional model of the bivalve mollusc sample.

[0070] like Figure 2 As shown, the three-dimensional model shooting device includes a shooting device 1, a lighting system 2, a 360° stage 4 and a background plate 5. The bivalve mollusc sample 3 is placed on the 360° stage 4. The 360° stage 4 is used to drive the shooting device to rotate 360° at a constant speed during shooting. The shooting device 1 is used to take pictures of one side, obliquely above and obliquely below the bivalve mollusc sample; the background plate 5 is used to ensure that the background of the taken picture is a pure black background to ensure the image quality; the position of the lighting system 2 is used to provide the light source required for shooting to improve the brightness of the shooting.

[0071] (201) Restore the shellfish shell obtained in step (1) and place it on a stage, turn on the stage power switch, and keep the stage rotating 360°;

[0072] (202) Turn on and adjust the camera 1, ensure that the lens is in focus, and place the camera 1 at a position 45° above the shell of the shellfish (i.e., the angle between the line connecting the camera 1 and the shell of the shellfish and the horizontal plane is 45°). Adjust the position of the background plate 5 to ensure that the background of the photograph is a pure black background. Rotate the stage 360° to collect photographs of different positions of the shell of the shellfish.

[0073] (203) The photographing device 1 is made horizontal with the shell of the shellfish (i.e., the angle between the line connecting the photographing device 1 and the shell of the shellfish and the horizontal plane is 0°), and the other steps are the same as (202). The stage is rotated 360° to collect photos of different positions of the shell of the shellfish;

[0074] (204) The photographing device 1 is positioned 45° below the shell of the shellfish (i.e., the angle between the line connecting the photographing device 1 and the shell of the shellfish and the horizontal plane is 45°), and the other steps are the same as (202). The stage is rotated 360° to collect photos of different positions of the shell of the shellfish;

[0075] (205) Collect the images taken in steps (202)-(204), process and model them using photogrammetry software (e.g., RealityCapture), and obtain a three-dimensional model (three-dimensional model file) of the sample, such as Figure 3 shown.

[0076] (3) Nine morphological indicators of the three-dimensional model are extracted by the bounding box method, namely length, width, height, length / width, height / length, width / height, ellipticity, sphericity and average curvature.

[0077] (301) Convert the 3D model file (such as OBJ file) into a corresponding point cloud file (such as PCD and TXT). Each vertex in the point cloud file includes 3D coordinates, color and other information. Use the Open3D library in Python to read the 3D model file and calculate the size of its axis-aligned bounding box. Then use the size of the bounding box as the length L, width W and height H of the 3D model, and calculate the ratios between them: length / width, height / length and width / height;

[0078] (302) Calculate the ellipticity E of the three-dimensional model, Wherein, L is the length of the three-dimensional model, H is the height of the three-dimensional model, and W is the width of the three-dimensional model;

[0079] (303) Calculate the sphericity of a three-dimensional model Where: S represents the surface area of ​​the three-dimensional model, and V represents the volume of the three-dimensional model;

[0080] (304) Calculate the average curvature H of the three-dimensional model. Use the TriMesh library to calculate the average curvature by calculating the curvature at each vertex of the three-dimensional model, and then average and normalize to obtain the curvature characteristics of each three-dimensional model. The curvature of vertex i is Mean curvature Among them, θ i represents the angle between the normals of the adjacent faces of vertex V, A i Represents the area of ​​adjacent faces.

[0081] (4) The above morphological indicators are integrated to obtain a morphological indicator data set. The species category is annotated after the corresponding data set according to the molecular identification results to complete the construction of the machine learning model data set. Then the data set is divided into a training set and a test set. The training set is input into the KNN model and the random forest model RF for training, and the test set is input into the trained KNN model and the random forest model RF for testing. The model training results are evaluated using classification accuracy.

[0082] Use RandomForestClassifier in the Jupyter Notebook platform to build a random forest classifier object. Among them, n_estimators: specifies the number of decision trees; max_depth: specifies the maximum depth of each subtree; oob_score: calculates the Out-Of-Bag error for random forest evaluation; random_state: sets the random seed to ensure the repeatability of the results.

[0083] The training set ratio was adjusted to 50%, 60%, 70%, 75%, 80%, 85% and 90% during the process. Figure 6 It can be seen that when the training set is 85%, KNN has the highest accuracy. Figure 7 It can be seen that under different proportions, the AUC value of RF is higher than that of KNN, and RF is better in distinguishing hidden species.

[0084] The importance of nine morphological indicators in the RF model was also analyzed. Figure 8 As shown in the figure, the importance of the nine morphological indicators from large to small is average curvature (Avgcurvature), width (Width), length (Length) / width (Width), width (Width) / height (Height), ellipticity (Ellipticity), length (Length), sphericity (Sphericity), height (Height) / length (Length), height (Height).

[0085] (5) According to steps (2)-(3), a morphological index data set of the shellfish sample to be tested is obtained and input into the trained KNN model or random forest model RF. The results show that it can accurately classify the shellfish sample. That is, the trained model can be used to classify bivalve shellfish samples (samples without molecular labels) that have not been identified using molecular technology.

[0086] Example 2

[0087] Steps (1)-(2) are the same as in Example 1.

[0088] (3) The 3D model file data of the shellfish sample in step (2) is labeled according to the species to which the shellfish sample belongs through previous molecular identification, and then the data is divided into a training set and a test set. The training set is input into the PointNet++ model to train the model, and the test set is input into the trained PointNet++ model to test it. The model training results are evaluated using classification accuracy.

[0089] The training set ratio was adjusted to 50%, 60%, 70%, 80% and 90% during the process. Fig. 9 It can be seen that when the training set is 80%, the PointNet++ model has the highest accuracy.

[0090] (4) According to step (2), the three-dimensional model file data of the shellfish sample to be tested is obtained and input into the trained PointNet++ model. The results show that it can accurately classify it. That is, the trained model can be used to classify bivalve shellfish samples that are not identified using molecular technology (samples without molecular tags).

Claims

1. A method for identifying bivalve molluscs by combining molecular identification with three-dimensional vision, characterized in that: The specific steps include: (1) Dissecting a bivalve mollusc sample, separating it into adductor muscle tissue and shell, extracting DNA from the adductor muscle tissue, amplifying the COI gene sequence by PCR and sequencing, and obtaining a molecular tag of the bivalve mollusc sample, wherein the molecular tag refers to the species to which the bivalve mollusc sample belongs; (2) restoring the shell of the bivalve mollusc obtained in step (1), taking photos of the shell of the bivalve mollusc at different angles using a three-dimensional model shooting device, and processing the photos to obtain a three-dimensional model of the bivalve mollusc sample; (3) The nine morphological indicators of the 3D model are extracted by the bounding box method, namely length L, width W, height H, length L / width W, height H / length L, width W / height H, ellipticity E, and sphericity and mean curvature H， The above morphological indices are integrated to obtain a morphological index data set; (4) According to the molecular identification results, the species category is annotated after the corresponding morphological index data set to complete the construction of the machine learning model data set. Then the data set is divided into a training set and a test set. The training set is input into the KNN model and the random forest model RF for training, and the test set is input into the trained KNN model and the random forest model RF for testing. The model training results are evaluated using classification accuracy; (5) According to steps (2)-(3), a morphological index dataset of the shellfish sample to be tested is obtained and input into the trained KNN model or random forest model RF, and the results show that it can accurately classify the shellfish sample; or (3') labeling the 3D model file data of the shellfish sample in step (2) according to the species to which the shellfish sample belongs according to the previous molecular identification, and then dividing the data into a training set and a test set, inputting the training set into the PointNet++ model to train the model, and inputting the test set into the trained PointNet++ model to test it, and evaluating the model training results using classification accuracy; (4') According to step (2), the three-dimensional model file data of the shellfish sample to be tested is obtained and input into the trained PointNet++ model. The results show that it can accurately classify the shellfish sample; Step (2) is as follows: (201) Restore the shellfish shell obtained in step (1) and place it on a stage, turn on the stage power switch, and keep the stage rotating 360°; (202) Turn on and adjust the camera to ensure that the lens is in focus, so that the camera is at a position 45° above the shell of the shellfish, adjust the position of the background plate to ensure that the background of the photograph is a pure black background, rotate the stage 360°, and collect photos of different positions of the shell of the shellfish; (203) The photographing device is made level with the shell of the shellfish, and the other steps are the same as (202), the stage is rotated 360 degrees, and photographs of different positions of the shell of the shellfish are collected; (204) The photographing device is positioned 45° below the shell of the shellfish, and the other steps are the same as (202). The stage is rotated 360° to collect photos of different positions of the shell of the shellfish; (205) collecting the images taken in steps (202)-(204), processing and modeling them using photogrammetry software, and obtaining a three-dimensional model of the sample; Step (3) is as follows: (301) Convert the 3D model file into a corresponding point cloud file. Each vertex in the point cloud file includes 3D coordinates and color information. Read the 3D model file through the Open3D library in Python and calculate the size of its axis-aligned bounding box. Then use the size of the bounding box as the length L, width W and height H of the 3D model, and calculate the ratios between them: length / width, height / length and width / height; (302) Calculate the ellipticity E of the three-dimensional model, , where L is the length of the three-dimensional model, H is the height of the three-dimensional model, and W is the width of the three-dimensional model; (303) Calculate the sphericity of a three-dimensional model , , where: S represents the surface area of ​​the three-dimensional model, V represents the volume of the three-dimensional model; (304) Calculate the average curvature of a three-dimensional model H， The TriMesh library is used to calculate the average curvature by calculating the curvature at each vertex of the 3D model, and then averaging and normalizing to obtain the curvature characteristics of each 3D model. Curvature , mean curvature ,in, represents the angle between the normals of the adjacent faces of vertex V, represents the area of ​​adjacent faces; The method is used for distinguishing between cryptic species of bivalve molluscs.

2. The method for identifying bivalve molluscs by combining molecular identification with three-dimensional vision according to claim 1, characterized in that: Step (1) is as follows: (101) The dissected adductor muscle tissue was placed in anhydrous ethanol solution for storage to prevent muscle tissue degradation; (102) Take 50 mg of adductor muscle tissue and place it in a sterile centrifuge tube. Add 500 μL of lysis buffer to the centrifuge tube, then add 5 μL of proteinase K. Mix by inverting the tube, and place it in a 40 °C water bath for digestion overnight. (103) A mixture of phenol, chloroform, and isoamyl alcohol in a volume ratio of 25:24:1 was added to the centrifuge tube, which was then placed in a rotating incubator for 20 minutes; (104) Place the centrifuge tube in a high-temperature refrigerated centrifuge and centrifuge at 12,000 rpm for 10 min at 4 °C; (105) Transfer the supernatant after centrifugation in step (104) to a new centrifuge tube, add 600 ml of isopropanol, and precipitate at -20°C for more than 30 minutes; (106) Centrifuge at 12,000 rpm for 10 min. A white precipitate is seen at the bottom. Discard the supernatant and keep the white precipitate at the bottom. (107) Add 75% ethanol solution to the white precipitate and centrifuge at 12,000 rpm for 10 min. Discard the supernatant and retain the precipitate. Repeat this process until a white to transparent precipitate is obtained, which is the DNA of the adductor muscle tissue. (108) Add 50 μL TE buffer to the precipitate obtained in the air-drying step (107) to dissolve the DNA, use a UV spectrophotometer to detect the OD value and concentration of the DNA solution, and use electrophoresis to detect the DNA solution; (109) The COI target fragment of the bivalve sample was amplified using a PCR amplification instrument. The reaction system was 25 μL, including 12.5 μL PCR mix premix, 10 μL H2O, 1 μL of upstream and downstream primers, and 0.5 μL template DNA. The upstream and downstream primer sequences were: LCO1490: 5′-GGTCAACAAATCATAAAGATATTGG-3′; HCO2198: 5-TAAACTTCAGGGTGACCAAAAAATCA-3′; (110) The amplified COI gene fragment of mitochondrial DNA is sequenced, and the sequencing results are compared with the NCBI database to determine the taxonomic status of the shellfish to be identified and obtain a molecular tag, which refers to the species to which the bivalve shellfish sample belongs.

3. The method for identifying bivalve molluscs by combining molecular identification with three-dimensional vision according to claim 1, characterized in that: The three-dimensional model shooting device in step (2) includes a shooting device, a lighting system, a 360° stage and a background plate. The bivalve mollusc sample is placed on the 360° stage. The 360° stage is used to drive the shooting device to rotate 360° at a constant speed during shooting. The shooting device is used to take photos of one side, obliquely above and obliquely below the bivalve mollusc sample; the background plate is used to ensure that the background of the photo is a pure black background to ensure image quality; the position of the lighting system is used to provide the light source required for shooting to improve the brightness of the shooting.

Images