A system for evaluating the health of fish fry

Abstract

The application discloses a kind of fry health degree evaluation system, including data acquisition module, model construction module and evaluation module, the data acquisition module includes activity analysis device and character analysis device, respectively determine the biological data and behavior data of fry, the model construction module is used to process the data collected by the data acquisition module, and utilize machine learning algorithm to construct prediction model;The evaluation module will utilize the prediction result output by the prediction model, according to fry health degree evaluation index, generate evaluation content.The application adopts above-mentioned a kind of fry health degree evaluation system, and fish fry video can be realized fry health evaluation by recording, so as to avoid the problem that fry is lost when contacting fry during fry detection;Meanwhile, a new index is defined as the judgment of fry health degree, to describe the degree of health state of fry.

Description

Technical Field

[0001] This invention relates to the field of fish fry health testing technology, and in particular to a fish fry health evaluation system. Background Technology

[0002] As fish fry grow, deformities and generally low survival rates occur. Therefore, improving the survival rate of fish fry plays an indispensable role in the development of fish fry farming.

[0003] With the development of computer image processing technology, image processing technology that can quickly calculate the total number of fish fry using computers has rapidly emerged. However, existing technologies cannot accurately identify traits such as size, color, damage, and activity of fish fry. Manual identification is time-consuming, labor-intensive, and has low accuracy, and may even damage the fish fry. Furthermore, it cannot identify large batches of fish fry continuously. Therefore, deep learning-based image processing methods can be used to identify fish fry traits. Summary of the Invention

[0004] The purpose of this invention is to provide a fish fry health evaluation system that improves the efficiency of fish fry detection; fish fry health evaluation can be achieved by recording fish fry videos, thereby avoiding the problem of fish fry loss caused by contact with fish fry during detection; by combining biological data and behavioral data, the problem of inaccurate prediction results caused by the one-sidedness of a single feature is avoided.

[0005] To achieve the above objectives, the present invention provides a fish fry health evaluation system, including a data acquisition module, a model building module, and an evaluation module. The data acquisition module includes an activity analysis device and a trait analysis device. The model building module is used to process the data acquired by the data acquisition module and to build a predictive model using machine learning algorithms. The evaluation module uses the prediction results output by the predictive model to generate evaluation content based on the fish fry health evaluation indicators. The model building module includes a model output submodule. The model output submodule verifies the accuracy of the prediction model based on the results of manually measured biological and behavioral data. If the accuracy reaches 90% or more, the model is output; if the accuracy does not reach 90% or more, training data is added and the model is trained again.

[0006] Preferably, the activity analysis device includes a first water tank, a first camera, a magnetic stirrer, and a water pump. The first camera is positioned above the first water tank, the magnetic stirrer is positioned below the first water tank, and lighting lamps are installed on both sides inside the first water tank. The water pump is positioned above the lighting lamps on one side of the first water tank and is connected to the water outlet below the first water tank via a water pipe. A baffle and a first valve are installed at the water outlet.

[0007] Preferably, the shape analysis device includes a second water tank and a second camera. The second water tank and the first water tank are connected by a water pipe, and a second valve is installed at both interfaces. The second camera is installed above the second water tank. The first camera and the second camera are connected to the same computer. The height of the second water tank is lower than that of the first water tank to facilitate shape observation by the second camera.

[0008] A method for evaluating the health of fish fry includes the following steps: S1. Use the first camera and the second camera to acquire videos and feature images of fish fry behavior; S2. A fish fry health model was built using a convolutional neural network as the main network. S3. Training a fish fry health model, with multiple features working together; S4. The trained fish fry health model evaluates the health status of the periodically input fish fry data in real time.

[0009] Preferably, the method for obtaining fish fry behavior videos and feature images in step S1 includes the following steps: S1-1, Activity Analysis Data Acquisition: Place the fish fry in the first water tank, turn on the lights, and use the first camera to record a video of the fish fry swimming normally; turn on the magnetic stirrer to create a vortex in the first water tank, and use the first camera to record the movement behavior of the fish fry in the vortex state; open the first valve, and use the water pump to simulate the behavior of the fish fry in the pouring state, and use the first camera to record a video, and upload the video of the fish fry behavior recorded by the first camera to the computer; S1-2, Shape Data Acquisition: Open the second valve, the fish fry flow into the second water tank, the second camera captures the surface features of the fish fry, and uploads the biological feature video to the computer; S1-3, Behavioral Data Preprocessing: The computer processes the behavioral video captured by the first camera into frames and performs image annotation. The behavioral dataset is expanded through data augmentation. Then, the behavioral data of fish fry movement in the video is analyzed through model training to obtain the final behavioral results. S1-4. Biological Data Preprocessing: The computer analyzes the body color and appearance of fish fry in the biological feature video captured by the second camera, extracting information on fry size, color, and damage. For fry size, a correspondence between body weight and body length is established using a calibration board and measurement data. For color and damage, RGB images captured by the second camera are input, and median filtering and histogram thresholding are used to remove noise and background interference other than fry. The obtained images are then labeled, with multiple labels set for body color and damage level to avoid insufficient dataset. Data augmentation is used for training to obtain a biological dataset, followed by model training to obtain the output results.

[0010] Preferably, the health of fish fry is measured using the fish fry health index. The (Fish Healthy Index) is calculated as follows: First, calculate the individual fish fry score D. Where m represents the number of evaluation items, This represents the score of the j-th evaluation item. This represents the weight coefficient of the j-th evaluation item; The individual fry scoring refers to scoring the fry's behavior, body color, and appearance. The behavior assessment items are divided into the fry's state under normal conditions, when agitated, and when the water is drained. The body color assessment items include color and gloss. The appearance assessment items include fish size, scales, fins, and skin. The scores for the assessment items are automatically obtained by the assessment module through deep learning, indicating the proportion of fry in the batch that are within the normal range for a certain assessment item. The weighting coefficients are obtained by expert scoring. Fish fry health score The calculation formula is as follows: The weights of the evaluation dimensions on the health of the fish fry are calculated based on their respective impacts. Where n represents the number of evaluation dimensions, This represents the score for the i-th dimension. This represents the weight coefficient of the i-th dimension; The assessment dimensions refer to behavior, body color, and appearance. The weight of the assessment dimensions on the health of the fry is obtained through expert scoring.

[0011] Therefore, the above-mentioned fish fry health evaluation system of the present invention has the following technical effects: (1) Fish fry health assessment can be achieved by recording videos of fish fry, thus avoiding the problem of fish fry loss caused by contact with fish fry during fish fry testing. (2) A health assessment method is provided, and the efficiency of fish fry testing is improved; (3) Combining biological and behavioral data can avoid the problem of inaccurate prediction results caused by the one-sidedness of a single feature.

[0012] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the activity analysis device and the shape analysis device in a fish fry health evaluation system of the present invention; Figure 2 This is a flowchart of a method for evaluating the health of fish fry according to the present invention; Figure 3 This is a flowchart of the preprocessing of biological data of fish fry in a fish fry health evaluation system of the present invention; Figure 4 This is a flowchart of the preprocessing of fish fry behavioral data in a fish fry health evaluation system of the present invention; Figure 5 This is a flowchart of the training and prediction model of the model output submodule in a fish fry health evaluation system of the present invention.

[0014] 1. First water tank; 2. First camera; 3. Lighting; 4. Magnetic stirrer; 5. First valve; 6. Water pump; 7. Water pipe; 8. Netting; 9. Second valve; 10. Pipeline; 11. Second water tank; 12. Second camera; 13. Computer. Detailed Implementation

[0015] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0017] Example 1 A fish fry health evaluation system includes a data acquisition module, a model building module, and an evaluation module. The data acquisition module includes an activity analysis device and a trait analysis device to measure the biological and behavioral data of the fish fry, respectively. The model building module processes the data acquired by the data acquisition module and uses machine learning algorithms to build a predictive model. The evaluation module uses the prediction results output by the predictive model to generate evaluation content based on the fish fry health evaluation indicators.

[0018] like Figure 5As shown, the model building module includes a model output submodule. The process of building the model output submodule includes training a prediction model using biological data, dividing the dataset into training and test sets, comparing and selecting the optimal network model, and incorporating transfer learning. The model is trained using the training set of the biological dataset, and then its accuracy is verified using the test set. If the accuracy reaches 90% or higher during verification, the model is output; otherwise, more training data is added. Then, the prediction results are corrected using the corresponding data from the biological dataset, thus obtaining the prediction model. Inputting test fish fry images outputs fish fry identification results, and multiple features are used to calculate the health status.

[0019] like Figure 1 As shown, the activity analysis device includes a first water tank 1, a first camera 2, a magnetic stirrer 4, and a water pump 6. The first camera 2 is located above the first water tank 1, the magnetic stirrer 4 is located below the first water tank 1, and lighting lamps 3 are installed on both sides inside the first water tank 1. The water pump 6 is located above the lighting lamp 3 on one side of the first water tank 1 and is connected to the water outlet below the first water tank 1 through a water pipe 7. A baffle 8 and a first valve 5 are installed at the water outlet.

[0020] The morphological analysis device includes a second water tank 11 and a second camera 12. The second water tank 11 and the first water tank 1 are connected by a pipe 10, and a second valve 9 is installed at both interfaces. The second camera 12 is installed above the second water tank 11. The first camera 2 and the second camera 12 are connected to the same computer 13. The height of the second water tank 11 is lower than that of the first water tank 1, which facilitates the second camera 12 to observe the morphological characteristics.

[0021] like Figure 2 As shown, the method for evaluating the health of fish fry includes the following steps: S1. Use the first camera and the second camera to acquire videos and feature images of fish fry behavior; S2. A fish fry health model was built using a convolutional neural network as the main network. S3. Training a fish fry health model, with multiple features working together; S4. The trained fish fry health model evaluates the health status of the periodically input fish fry data in real time.

[0022] The method for obtaining fish fry behavior videos and feature images in step S1 includes the following steps: S1-1, Activity Analysis Data Acquisition: Place the fish fry in the first water tank 1, turn on the light 3, and use the first camera 2 to record a video of the fish fry swimming normally; turn on the magnetic stirrer 4 to generate a vortex in the first water tank 1, and use the first camera 2 to record the movement behavior of the fish fry in the vortex state; open the first valve 5, and use the water pump 6 to simulate the behavior of the fish fry in the pouring state, and use the first camera 2 to record a video, and upload the video of the fish fry behavior recorded by the first camera 2 to the computer 13.

[0023] S1-2, Trait Data Collection: Open the second valve 9, the fish fry flow into the second water tank 11, the second camera 12 captures the surface features of the fish fry, and uploads the biological feature video to the computer 13; S1-3, Behavioral data preprocessing: such as... Figure 3 As shown, the computer processes the behavioral video captured by the first camera into frames, performs image annotation, expands the behavioral dataset through data augmentation, and then analyzes the behavioral data of fish fry movement in the video through model training to obtain the final behavioral results. S1-4. Biological data preprocessing: such as... Figure 4 As shown, the computer analyzes the body color and appearance of fish fry in the biometric video captured by the second camera, extracting information on the size, color, and damage of the fry. For the size of the fry, a correspondence between body weight and body length is established using a calibration board and measurement data. For the color and damage, RGB images captured by the second camera are input, and median filtering and histogram thresholding are used to remove noise and background interference other than the fry. The obtained images are then labeled, and multiple labels can be set for body color and damage to avoid the problem of insufficient dataset. Data augmentation is used for training to obtain a biological dataset, and then the model is trained to obtain the output results.

[0024] This invention involves capturing behavioral and biological characteristic videos of fish fry, transmitting the videos to a computer, and then using a deep learning-based computer to preprocess the videos, analyze the behavior and traits of the fish fry in the videos, and finally compare them with a pre-defined fish fry grading system to analyze the health of the fish fry.

[0025] Fish fry health is measured by the fish fry health index. The (Fish Healthy Index) is calculated as follows: First, the individual fish fry score D is calculated using an additive model, and the calculation formula is as follows: Where m represents the number of evaluation items, This represents the score of the j-th evaluation item. This represents the weight coefficient of the j-th evaluation item; The individual fry scoring refers to rating the fry's behavior, body color, and appearance. Behavior is assessed in three ways: normal state, state during agitation, and state after draining water. Body color assessment includes color and gloss. Appearance assessment includes body size, scales, fins, and skin. The scores for each assessment item are automatically generated by the assessment module through deep learning, representing the proportion of fry in the batch that fall within the high-quality range for a given assessment item. Weighting coefficients are obtained using an expert scoring method. The characteristics of each assessment item for the fry are shown in Table 1.

[0026] Fish fry health score The calculation formula is as follows: The weights of the evaluation dimensions on the health of the fish fry are calculated based on their respective impacts. Where n represents the number of evaluation dimensions, This represents the score for the i-th dimension. This represents the weight coefficient of the i-th dimension; The assessment dimensions refer to behavior, body color, and appearance. The weight of the assessment dimensions on the health of the fry is obtained through expert scoring.

[0027] Table 1 Performance of Fish Fry in Each Assessment Item

[0028] Example 2 Unlike the method for training the model in the model output submodule in Example 1, this method includes: dividing the behavior dataset into training and test sets, comparing and selecting the optimal network model, and incorporating transfer learning. The model is trained using the training set of the behavior dataset, its accuracy is verified using the test set, and then the prediction results are corrected using the corresponding biological data to obtain the prediction model.

[0029] The model training results consist of behavioral data, including parameters such as acceleration, velocity, distance, curvature, angular velocity, and trajectory. A clustering analysis algorithm is used to obtain energy cluster centers for each waveband. The distance between the energy cluster centers of the data to be predicted and the obtained energy cluster centers for each health level is used to check the similarity of the detected data. This similarity is then compared with expert standards to obtain the prediction result. The expert standards are evaluation criteria derived from experts' classification of fish fry health levels.

[0030] Therefore, the present invention provides a fish fry health evaluation system with the above-mentioned structure, which can evaluate the health of fish fry by recording fish fry videos, thereby avoiding the problem of fish fry loss caused by contact with fish fry during fish fry detection. At the same time, it provides a health evaluation method that combines biological data and behavioral data to avoid the problem of inaccurate prediction results caused by the one-sidedness of a single feature, and improves the efficiency of fish fry detection.

[0031] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A fish fry health evaluation system, characterized in that: It includes a data acquisition module, a model building module, and an evaluation module. The data acquisition module includes an activity analysis device and a trait analysis device to measure the biological and behavioral data of the fish fry, respectively. The model building module is used to process the data collected by the data acquisition module and build a predictive model using machine learning algorithms. The evaluation module will use the prediction results output by the predictive model to generate evaluation content based on the fish fry health evaluation indicators. The model building module includes a model output submodule. The model output submodule verifies the accuracy of the prediction model based on the results of manually measured biological and behavioral data. If the accuracy reaches 90% or more, the model is output; if the accuracy does not reach 90% or more, training data is added and the model is trained again. The activity analysis device includes a first water tank, a first camera, a magnetic stirrer, and a water pump. The first camera is located above the first water tank, the magnetic stirrer is located below the first water tank, and lighting lamps are installed on both sides inside the first water tank. The water pump is located above the lighting lamp on one side of the first water tank and is connected to the water outlet below the first water tank through a water pipe. A baffle and a first valve are installed at the water outlet. The shape analysis device includes a second water tank and a second camera. The second water tank and the first water tank are connected by a pipe, and a second valve is installed at both interfaces. The second camera is installed above the second water tank. The first camera and the second camera are connected to the same computer. The second water tank is lower than the first water tank to facilitate shape observation by the second camera.

2. A method for evaluating the health of fish fry using the fish fry health evaluation system as described in claim 1, characterized in that, Includes the following steps: S1. Use the first camera and the second camera to acquire videos and feature images of fish fry behavior; S2. A fish fry health model was built using a convolutional neural network as the main network. S3. Training a fish fry health model, with multiple features working together; S4. The trained fish fry health model evaluates the health status of the periodically input fish fry data in real time.

3. The method for evaluating the health of fish fry according to claim 2, characterized in that, The method for obtaining fish fry behavior videos and feature images in step S1 includes the following steps: S1-1, Activity Analysis Data Acquisition: Place the fish fry in the first water tank, turn on the lights, and use the first camera to record video of the fish fry swimming normally; turn on the magnetic stirrer to create a vortex in the first water tank, and use the first camera to record the movement behavior of the fish fry in the vortex state. Open the first valve, use the water pump to simulate the behavior of fish fry under the condition of pouring water, and use the first camera to shoot video. Upload the video of the fish fry behavior shot by the first camera to the computer. S1-2, Trait Data Collection: Open the second valve, the fish fry flow into the second water tank, the second camera captures the surface features of the fish fry, and the biological feature video is uploaded to the computer; S1-3, Behavioral Data Preprocessing: The computer processes the behavioral video captured by the first camera into frames and performs image annotation. The behavioral dataset is expanded through data augmentation. Then, the behavioral data of fish fry movement in the video is analyzed through model training to obtain the final behavioral results. S1-4. Biological Data Preprocessing: The computer analyzes the body color and appearance of fish fry in the biological feature video captured by the second camera, extracting information on fry size, color, and damage. For fry size, a correspondence between body weight and body length is established using a calibration board and measurement data. For color and damage, RGB images captured by the second camera are input, and median filtering and histogram thresholding are used to remove noise and background interference other than fry. The obtained images are then labeled, with multiple labels set for body color and damage level to avoid insufficient dataset. Data augmentation is used for training to obtain a biological dataset, followed by model training to obtain the output results.

4. The method for evaluating the health of fish fry according to claim 3, characterized in that, Fish fry health is measured by the fish fry health index. The calculation method is as follows: First, calculate the individual fish fry score D using an additive model. The calculation formula is as follows: Where m represents the number of evaluation items, This represents the score of the j-th evaluation item. This represents the weight coefficient of the j-th evaluation item; The individual fry scoring refers to scoring the fry's behavior, body color, and appearance. The behavior assessment items are divided into the fry's state under normal conditions, when agitated, and when the water is drained. The body color assessment items include color and gloss. The appearance assessment items include fish size, scales, fins, and skin. The scores for the assessment items are automatically obtained by the assessment module through deep learning, indicating the proportion of fry in the same batch that are within the normal range for a certain assessment item. The weighting coefficients are obtained by expert scoring. Fish fry health score The calculation formula is as follows: The weights of the evaluation dimensions on the health of the fish fry are calculated based on their respective impacts. Where n represents the number of evaluation dimensions, This represents the score for the i-th dimension. This represents the weight coefficient of the i-th dimension; The assessment dimensions refer to behavior, body color, and appearance. The weight of the assessment dimensions on the health of the fry is obtained through expert scoring.

Images