Underwater fish body mass measurement method and system

Abstract

The present invention discloses an underwater fish body mass measurement method and system, comprising: obtaining two left and right fish body images, namely a left fish body image and a right fish body image; obtaining the key points of the fish from the left fish body image and the right fish body image; obtaining the sparse depth information of the fish at the key points according to the key points of the fish; quantifying the correlation between each morphological parameter and the fish body mass according to the Garson algorithm, in combination with the voting algorithm and the randomized experiment; selecting the morphological parameter according to the quantified result of the correlation between each morphological parameter and the fish body mass; obtaining the selected fish body morphological parameter information according to the sparse depth information; obtaining the fish body mass according to the selected fish body morphological parameter information and the trained fish body mass prediction model. The method can measure the fish body mass quickly and non-destructively.

Description

Technical Field

[0001] The present invention relates to the technical field of fish mass measurement, and in particular to an underwater fish mass measurement method and system. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Fish body mass is an important indicator that determines fish production performance and is also one of the most direct breeding goals. In various fish farming facilities, fish body mass is usually obtained by manual weighing after sampling fish. The detection process requires a lot of manpower and material resources and is inefficient. The measurement accuracy is limited by subjective factors such as the experience and habits of the tester, and it is impossible to form a unified measurement standard. In order to prevent the measured individuals from having a stress response, they need to be dehydrated for a long time, which causes physical damage to farmed fish, especially fry, which has a certain impact on subsequent research and also causes economic losses to fish farmers. Summary of the invention

[0004] In order to solve the above problems, the present disclosure proposes an underwater fish mass measurement method and system, which can measure the fish mass quickly and non-destructively.

[0005] To achieve the above objectives, the present disclosure adopts the following technical solutions:

[0006] In the first aspect, a method for measuring the underwater fish mass is proposed, comprising:

[0007] Obtain two fish body images, namely, a left fish body image and a right fish body image;

[0008] Obtain the key points of the fish from the left and right images of the fish body;

[0009] Obtain sparse depth information of the fish at the key points according to the key points of the fish;

[0010] The selected fish body morphological parameter information is obtained according to the sparse depth information, wherein the process of obtaining the selected fish body morphological parameters is as follows: obtaining the morphological parameters of the fish body, constructing a neural network model with the morphological parameters of the fish body as input and the fish body mass as output, calculating the contribution matrix of the neural network model, converting the contribution matrix into a ranking matrix, calculating the confidence of the morphological parameters of the fish body according to the ranking matrix, and obtaining a reasonable ranking matrix according to the confidence; screening the neural network model according to the reasonable ranking matrix, and setting initial weights for the screened neural network model; performing a randomized experiment with the screened neural network model and the initial weights as input to obtain the significance of the contribution; determining the selected fish body morphological parameters according to the significance;

[0011] The fish body mass is obtained based on the morphological parameter information of the fish body and the trained fish body mass prediction model.

[0012] In the second aspect, an underwater fish mass measurement system is proposed, comprising:

[0013] An image acquisition module is used to acquire two left and right fish body images, namely a left fish body image and a right fish body image;

[0014] A key point extraction module is used to obtain the key points of the fish from the left image and the right image of the fish body;

[0015] A sparse depth information acquisition module, used to acquire sparse depth information of the fish at key points according to the key points of the fish;

[0016] The fish body morphological parameter acquisition module is used to obtain the selected fish body morphological parameter information according to the sparse depth information, wherein the process of obtaining the selected fish body morphological parameters is as follows: obtaining the morphological parameters of the fish body, constructing a neural network model with the morphological parameters of the fish body as input and the fish body mass as output, calculating the contribution matrix of the neural network model, converting the contribution matrix into a ranking matrix, calculating the confidence of the morphological parameters of the fish body according to the ranking matrix, and obtaining a reasonable ranking matrix according to the confidence; screening the neural network model according to the reasonable ranking matrix, and setting the initial weight for the screened neural network model; performing a randomized experiment with the screened neural network model and the initial weight as input to obtain the significance of the contribution; determining the selected fish body morphological parameters according to the significance;

[0017] The fish body mass acquisition module is used to obtain the fish body mass based on the morphological parameter information of the fish body and the trained fish body mass prediction model.

[0018] In a third aspect, an electronic device is proposed, comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps described in a method for measuring underwater fish mass are completed.

[0019] In a fourth aspect, a computer-readable storage medium is proposed for storing computer instructions, which, when executed by a processor, complete the steps described in a method for measuring underwater fish mass.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] 1. The present invention estimates fish mass based on the analysis results of the correlation between fish morphological parameters and fish mass, and can measure fish mass quickly and non-destructively.

[0022] 2. In order to accurately measure the weight of fish, the present invention selects the morphological parameters of the fish according to the correlation between the morphological parameters of the fish and the weight of the fish. When the weight of the fish is estimated by the selected morphological parameters of the fish, the estimated weight of the fish is more accurate.

[0023] 3. When selecting the morphological parameters of the fish body, the present invention, based on the neural network input contribution estimation algorithm, combines the voting method and the randomized experiment to quantify the correlation between the fish body morphological parameters and the fish body quality, and can more accurately select the fish body morphological parameters. On this basis, the measurement accuracy of the fish body quality is effectively guaranteed.

[0024] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0026] Figure 1 is a flow chart of the method disclosed in Example 1;

[0027] Figure 2 The quantitative analysis process of the correlation between the morphological parameters of the fish body and the fish body quality disclosed in Example 1;

[0028] Figure 3 The extreme learning machine model disclosed in Example 1;

[0029] Figure 4 The flowchart of the chaotic particle swarm optimization algorithm disclosed in Example 1;

[0030] Figure 5 The key point positioning flow chart of the fish disclosed in Example 1;

[0031] Figure 6 The morphological parameters of the fish disclosed in Example 1;

[0032] Figure 7 is the shape vector of the fish disclosed in Example 1;

[0033] Figure 8 The key points of the fish body located in Example 1;

[0034] Fig. 9 It is the key point sparse depth map disclosed in Example 1.

[0035] Among them: ①, total length, ②, body length, ③, body height, ④, eye diameter, ⑤, head length, ⑥, snout length, ⑦, caudal peduncle length. DETAILED DESCRIPTION

[0036] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0037] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0038] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0039] Example 1

[0040] In order to accurately measure the fish body mass, an underwater fish body mass measurement method is disclosed in this embodiment, including:

[0041] Obtain two fish body images, namely, a left fish body image and a right fish body image;

[0042] Obtain the key points of the fish from the left image and the right image of the fish body;

[0043] Obtain sparse depth information of the fish at the key points according to the key points of the fish;

[0044] The selected fish body morphological parameter information is obtained according to the sparse depth information, wherein the process of obtaining the selected fish body morphological parameters is as follows: obtaining the morphological parameters of the fish body, constructing a neural network model with the morphological parameters of the fish body as input and the fish body mass as output, calculating the contribution matrix of the neural network model, converting the contribution matrix into a ranking matrix, calculating the confidence of the morphological parameters of the fish body according to the ranking matrix, and obtaining a reasonable ranking matrix according to the confidence; screening the neural network model according to the reasonable ranking matrix, and setting initial weights for the screened neural network model; performing a randomized experiment with the screened neural network model and the initial weights as input to obtain the significance of the contribution; determining the selected fish body morphological parameters according to the significance;

[0045] The fish body mass is obtained based on the morphological parameter information of the fish body and the trained fish body mass prediction model.

[0046] Furthermore, the contribution matrix of the morphological parameters of the fish body is calculated by the Garson algorithm.

[0047] Furthermore, the specific process of calculating the confidence of the morphological parameters of the fish body according to the ranking matrix is ​​as follows:

[0048] Count the votes for each position in the ranking matrix;

[0049] A two-dimensional contingency table was constructed based on the morphological parameters of the fish and the corresponding votes;

[0050] The confidence levels of the morphological parameters of the fish were calculated based on the two-dimensional contingency table.

[0051] Furthermore, the fish quality prediction model and the neural network model both adopt the extreme learning machine model.

[0052] Furthermore, chaotic particle swarm optimization is used to optimize the extreme learning machine model.

[0053] Furthermore, a randomized experiment is conducted with the selected network model and initial weight as input, and the specific process of obtaining the significance of contribution is as follows:

[0054] Randomly change the output order of training samples to train the selected network model, calculate the randomized contribution, and convert the randomized contribution into the current ranking matrix;

[0055] The sample data is screened by the similarity between the current ranking matrix and the reasonable ranking matrix;

[0056] Calculate the significance of randomization contribution through the selected sample data;

[0057] The significance of the randomized contributions is averaged to obtain the final significance of the contribution.

[0058] Furthermore, the spatial three-dimensional coordinates of the key points are calculated through the sparse depth information of the key points, the distances between the key points are calculated according to the spatial three-dimensional coordinates, and the selected fish body morphological parameter information is determined through the distances between the key points.

[0059] A method for measuring the underwater fish mass disclosed in this embodiment is discussed in detail.

[0060] A method for measuring the mass of underwater fish, such as Figure 1 As shown, it includes the first and second stages.

[0061] The first stage is the preparation stage of the fish quality prediction model, which includes:

[0062] like Figure 2 As shown, the morphological parameters of the fish body are selected according to the correlation between the morphological parameters of the fish body and the quality of the fish body, so as to obtain the selected morphological parameters of the fish body;

[0063] A fish body quality prediction model is constructed according to the selected fish body morphological parameter information, and the constructed fish body quality prediction model is trained to obtain a trained fish body quality prediction model.

[0064] The specific process of obtaining the selected fish body morphological parameters is as follows:

[0065] S11: Acquire morphological parameter information of the fish body, and construct training samples of the neural network model through the morphological parameter information of the fish body.

[0066] S12: Based on the acquired morphological parameter information of the fish body, a neural network model is constructed with the morphological parameters of the fish body as input and the fish body mass as output, and a contribution matrix of the neural network model is calculated.

[0067] In the specific implementation, t neural network models are constructed with the morphological parameters of the fish body as input and the weight of the fish body as output. When the number of the determined morphological parameters of the fish body is g, the number of input variables of the neural network model is g.

[0068] The contribution matrix of the fish body morphological parameters to the fish body mass is obtained by the Garson algorithm: U = {u 1 ,u 2 ,…,u t}, the flowchart of Garson algorithm is as follows Figure 3 shown.

[0069] S13: Convert the contribution matrix into a ranking matrix.

[0070] In specific implementation, the contribution matrix U is transformed into a ranking matrix P = {p 1 , p 2 , ..., p t}∈R g : Replace u with 1~g 1 ,u 2 ,…,u t The contribution value in , the largest contribution value is 1, the second is 2, and so on, until the smallest contribution value is replaced by g.

[0071] S14: Calculate the confidence of the morphological parameters of the fish body according to the ranking matrix.

[0072] In the specific implementation, the ranking matrix P is transformed into a two-dimensional contingency table C with r rows and c columns: the individuals in the population t can be classified according to the two attributes of input variable A and ranking position B, where the input variable A is the morphological parameter of the fish body, and A has c levels A 1 ,A 2 ,…,A c , B has c levels B 1 ,B 2 ,…,B c. There are t samples in the population, assuming that there are t ij The attributes of each individual belong to level A i and B j , n ij It is called frequency, and r×c n ij Arranged as a two-dimensional contingency table with r rows and c columns, referred to as an r×c table.

[0073] Calculate the confidence of the morphological parameters of the input variable fish body. The confidence is calculated as:

[0074] Among them, C i is the ranking vector of the i-th input variable.

[0075] According to the confidence of the obtained fish body morphological parameters, the ranking rationality matrix is ​​obtained

[0076] S15: Filter the neural network model according to the reasonable ranking matrix, and set the initial weight for the filtered neural network model; perform a randomized experiment with the filtered network model and the initial weight as input to obtain the significance of the contribution; determine the selected fish body morphology parameters according to the significance of the contribution.

[0077] include:

[0078] S151: According to the ranking reasonable matrix Select m neural network models with reasonable rankings from the t neural network models, and record the initial weights of the m neural network models.

[0079] S152: Using m neural network models and initial weights corresponding to the models as inputs to conduct random experiments to obtain the significance of contribution, including:

[0080] S1521: Train the m neural network models through the training sample set, obtain the termination weights of the m neural network models, and calculate the contribution RI according to the termination weights.

[0081] S1522: Randomly change the order of the training sample output set, retrain the neural network model using the samples after the order is changed and the recorded initial weights of the m screened neural network models, and record the termination weight of the neural network model.

[0082] S1522 is repeated a large number of times, and the number of repetitions is recorded with COUNT. Each time, the randomized contribution RI is obtained using the Garson algorithm based on the termination weight recorded in S1522, and the contribution obtained by the termination weight corresponding to the neural network model with the best prediction performance is selected as the standard value of the randomized contribution RI.

[0083] According to the method in S13, the randomized contribution matrix RI obtained in S1521 and S1522 is converted into the current ranking vector Calculated by distance metric function and According to the similarity, the sample data is screened to remove the data with small similarity and retain the data with large similarity.

[0084] The significance level P of RI is calculated based on the screened data. The significance level is the probability of the statistical test. The calculation formula of the significance level P is as follows: 1) If the RI value is greater than 0, then P = (N+1) / (COUNT+1), where N is the number of RIs greater than or equal to the standard value; 2) If the RI value is less than 0, then P = (m+1) / (COUNT+1), where M is the number of RIs less than or equal to the standard value.

[0085] The significance level P of the m groups of RI corresponding to the m neural network models is then averaged as the final significance level.

[0086] S153: Eliminate redundant fish body morphological parameters according to the final significance level, select fish body morphological parameters from the fish body morphological parameters, and obtain selected fish body morphological parameters.

[0087] Through Figure 4 The extreme learning machine shown in the figure constructs a fish quality prediction model. Figure 5 The chaotic particle swarm shown optimizes the extreme learning machine, and uses the fish morphology parameters selected in S152 as input to set a training set, trains the constructed fish quality prediction model, and obtains a trained fish quality prediction model.

[0088] The second stage: real-time fish measurement stage, including:

[0089] S21: Acquire two left and right images of the fish body, namely a left fish body image and a right fish body image.

[0090] In a specific implementation, two calibrated cameras are used to obtain two corrected fish body images, namely a left fish body image and a right fish body image, after extreme transformation.

[0091] S22: Using the active shape model, key points of the left image and the right image of the fish body are located to obtain key points of the fish.

[0092] S23: Combined with the binocular depth measurement principle, the key points of the fish are analyzed to obtain the sparse depth information of the fish at the key points.

[0093] S24: Calculate the spatial three-dimensional coordinates of the key points through the sparse depth information of the key points, and then calculate the distance between each key point, and determine the value of the selected fish body morphology parameter through the distance between each key point, that is, the selected fish body morphology parameter information.

[0094] S25: Obtaining the fish body mass according to the obtained selected fish body morphological parameter information and the trained fish body mass prediction model.

[0095] In specific implementation, the selected fish body morphology parameter information is input into a trained fish body quality prediction model, and the fish body quality is calculated based on the mapping relationship between the fish body morphology parameters and the fish body quality in the trained fish body quality prediction model.

[0096] A method for measuring the underwater fish mass disclosed in this embodiment is described in detail by taking a 4-month-old greenfin pufferfish as an example.

[0097] like Figure 7 As shown in the figure, the morphological parameters directly or indirectly related to body mass of 4-month-old greenfin pufferfish include total length①, body length②, body height③, eye diameter④, head length⑤, snout length⑥, caudal peduncle length⑦, etc. The definitions of these 7 morphological parameters are shown in Table 1.

[0098] Table 1 Morphological parameters

[0099]

[0100] Ten key points are defined to describe the phenotypic characteristics of fish. The definition of key points is shown in Table 2, which contains the key point number and the location of the key point. Figure 8 The intuitive position of the key points on the fish body and the outline of its shape vector are shown.

[0101] Table 2 Key points

[0102]

[0103]

[0104] Generate 500 extreme learning machines (ELMs) with random initial weights, train them, and calculate the output contribution respectively, and you will get a contribution matrix consisting of 500 contribution vectors.

[0105] The contribution matrix is ​​converted into a ranking variable matrix, and the number of votes for each position in the ranking variable matrix is ​​counted to create a contingency table. The contingency table is shown in Table 3.

[0106] Table 3 Contingency table

[0107]

[0108] After calculating the confidence according to the contingency table, the reasonable ranking vector was obtained: body length, total length, body height, head length, dorsal fin distance, and eye diameter.

[0109] 305 ELM networks were selected based on the ranked reasonable vectors, the initial weights were fixed, and after training again and calculating the contribution, the initial weight matrix was recorded.

[0110] Randomly change the order of the training set and retrain from the initial weight matrix, and iterate 100 times.

[0111] For each ELM network, 100 groups of contribution vectors RI were finally recorded. According to the Pearson correlation coefficient, the data that did not meet the ranking were removed, and the remaining data were used to obtain the significance of contribution according to the significance formula of the randomized experiment. In this way, a total of 305 groups of significance were obtained. The significance of the morphological parameters of each input variable fish body was averaged as the final significance result. The final results are shown in Table 4.

[0112] Table 4 Significance index of input index contribution

[0113]

[0114] The eye diameter morphological parameter was discarded to train the fish mass prediction model, and a trained fish mass prediction model was obtained.

[0115] Obtain the left image and the right image of the fish body, correct the two images by using epipolar transformation, and locate the key points of the corrected left and right images. The located key points are as follows: Figure 8 shown.

[0116] The disparity values ​​of the 10 key points are calculated to obtain the sparse depth information of the fish at the key points, such as Fig. 9 The coordinates of some key points and the obtained disparity values ​​are shown in Table 5.

[0117] Table 5 Coordinates of some key points

[0118]

[0119] The distances between key points are calculated and the three-dimensional geometric features of the fish body are extracted to complete the determination of the morphological parameters. The obtained morphological parameter determination results are shown in Table 6.

[0120] Table 6 Morphological measurement results

[0121]

[0122]

[0123] The obtained morphological parameters are input into the trained fish body mass prediction model to obtain the fish body mass.

[0124] The method disclosed in this embodiment estimates the fish mass based on the analysis results of the correlation between the fish morphological parameters and its body mass, and can measure the fish mass quickly and non-destructively.

[0125] In order to accurately measure the fish mass, this embodiment selects the morphological parameters of the fish body according to the correlation between the morphological parameters of the fish body and the fish mass. When the fish mass is estimated by using the selected fish morphological parameters, the estimated fish mass is more accurate.

[0126] When selecting the morphological parameters of the fish body, this embodiment, based on the neural network input contribution estimation algorithm, combines the voting method and the randomized experiment to quantify the correlation between the fish body morphological parameters and the fish body quality, and can more accurately select the fish body morphological parameters. On this basis, the measurement accuracy of the fish body quality is effectively guaranteed.

[0127] Example 2

[0128] In this embodiment, an underwater fish mass measurement system is disclosed, comprising:

[0129] An image acquisition module is used to acquire two left and right fish body images, namely a left fish body image and a right fish body image;

[0130] A key point extraction module is used to obtain the key points of the fish from the left image and the right image of the fish body;

[0131] A sparse depth information acquisition module, used to acquire sparse depth information of the fish at key points according to the key points of the fish;

[0132] The fish body morphological parameter acquisition module is used to acquire the selected fish body morphological parameters according to the sparse depth information, wherein the process of acquiring the selected fish body morphological parameters is as follows: acquiring the morphological parameters of the fish body, constructing a neural network model with the morphological parameters of the fish body as input and the fish body mass as output, calculating the contribution matrix of the neural network model, converting the contribution matrix into a ranking matrix, calculating the confidence of the morphological parameters of the fish body according to the ranking matrix, and obtaining a reasonable ranking matrix according to the confidence; screening the neural network model according to the reasonable ranking matrix, and setting initial weights for the screened neural network model; performing randomized experiments with the screened neural network model and the initial weights as input to obtain the significance of the contribution; determining the selected fish body morphological parameters according to the significance;

[0133] The fish body mass acquisition module is used to obtain the fish body mass according to the morphological parameters of the fish body and the trained fish body mass prediction model.

[0134] Example 3

[0135] In this embodiment, an electronic device is disclosed, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the steps described in the underwater fish body mass measurement method disclosed in Example 1 are completed.

[0136] Example 4

[0137] In this embodiment, a computer-readable storage medium is disclosed for storing computer instructions. When the computer instructions are executed by a processor, the steps described in the underwater fish body mass measurement method disclosed in Example 1 are completed.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for measuring the mass of underwater fish, characterized in that: include: Obtain two fish body images, namely, a left fish body image and a right fish body image; Obtain the key points of the fish from the left image and the right image of the fish body; Obtain sparse depth information of the fish at the key points according to the key points of the fish; The selected fish body morphological parameter information is obtained according to the sparse depth information, wherein the process of obtaining the selected fish body morphological parameters is as follows: obtaining the morphological parameters of the fish body, constructing a neural network model with the morphological parameters of the fish body as input and the fish body mass as output, calculating the contribution matrix of the neural network model, converting the contribution matrix into a ranking matrix, calculating the confidence of the morphological parameters of the fish body according to the ranking matrix, and obtaining a reasonable ranking matrix according to the confidence; screening the neural network model according to the reasonable ranking matrix, and setting initial weights for the screened neural network model; performing a randomized experiment with the screened neural network model and the initial weights as input to obtain the significance of the contribution; determining the selected fish body morphological parameters according to the significance; Obtaining the fish body mass according to the selected fish body morphological parameter information and the trained fish body mass prediction model; The specific process of calculating the confidence of the morphological parameters of the fish body according to the ranking matrix is: Count the votes for each position in the ranking matrix; A two-dimensional contingency table was constructed based on the morphological parameters of the fish and the corresponding votes; The confidence of the morphological parameters of the fish body was calculated based on the two-dimensional contingency table; The specific process of randomizing the experiment with the selected network model and initial weight as input to obtain the significance of contribution is as follows: Randomly change the output order of training samples to train the selected network model, calculate the randomized contribution, and convert the randomized contribution into the current ranking matrix; The sample data is screened by the similarity between the current ranking matrix and the reasonable ranking matrix; Calculate the significance of randomization contribution through the selected sample data; The significance of the randomized contributions is averaged to obtain the final significance of the contribution.

2. The underwater fish mass measurement method according to claim 1, characterized in that: The contribution matrix of the morphological parameters of the fish body to the fish body mass in the neural network model is calculated using the Garson algorithm.

3. The underwater fish mass measurement method according to claim 1, characterized in that: Both the neural network model and the fish weight prediction model use the extreme learning machine model.

4. A method for measuring underwater fish mass as claimed in claim 3, characterized in that: Chaotic particle swarm optimization is used to optimize the extreme learning machine model.

5. The underwater fish mass measurement method according to claim 1, characterized in that: The spatial three-dimensional coordinates of the key points are calculated through the sparse depth information of the key points, the distances between the key points are calculated according to the spatial three-dimensional coordinates, and the selected fish body morphological parameter information is determined through the distances between the key points.

6. An underwater fish mass measurement system, characterized in that: include: An image acquisition module is used to acquire two left and right fish body images, namely a left fish body image and a right fish body image; A key point extraction module is used to obtain the key points of the fish from the left image and the right image of the fish body; A sparse depth information acquisition module, used to acquire sparse depth information of the fish at key points according to the key points of the fish; The fish body morphological parameter acquisition module is used to obtain the selected fish body morphological parameters according to the sparse depth information, wherein the process of obtaining the selected fish body morphological parameters is as follows: obtaining the morphological parameters of the fish body, constructing a neural network model with the morphological parameters of the fish body as input and the fish body mass as output, calculating the contribution matrix of the neural network model, converting the contribution matrix into a ranking matrix, calculating the confidence of the morphological parameters of the fish body according to the ranking matrix, and obtaining a reasonable ranking matrix according to the confidence; screening the neural network model according to the reasonable ranking matrix, and setting initial weights for the screened neural network model; performing randomized experiments with the screened neural network model and the initial weights as input to obtain the significance of the contribution; determining the selected fish body morphological parameters according to the significance; calculating the fish body according to the ranking matrix The specific process of the confidence of the morphological parameters is as follows: count the number of votes for each position in the ranking matrix; construct a two-dimensional contingency table based on the morphological parameters of the fish body and the corresponding number of votes; calculate the confidence of the morphological parameters of the fish body based on the two-dimensional contingency table; conduct a randomized experiment with the selected network model and the initial weight as input, and the specific process of obtaining the significance of the contribution is as follows: randomly change the output order of the training samples to train the selected network model, calculate the randomized contribution, and convert the randomized contribution into the current ranking matrix; screen the sample data by the similarity between the current ranking matrix and the reasonable ranking matrix; calculate the significance of the randomized contribution through the screened sample data; average the significance of the randomized contribution to obtain the significance of the final contribution; The fish body mass acquisition module is used to obtain the fish body mass according to the selected fish body morphological parameters and the trained fish body mass prediction model.

7. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the underwater fish body mass measurement method according to any one of claims 1 to 5 are completed.

8. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of an underwater fish mass measurement method as described in any one of claims 1-5.

Images