Precise feeding system and method based on machine vision

Abstract

The invention discloses an accurate feeding system and method based on machine vision. The system comprises an image acquisition module, an image processing module, a satiation degree evaluation module, an automatic feeding machine and a control module. The image acquisition module is used for acquiring a water surface spray image of the fishpond in real time; the image processing module is in signal connection with the image acquisition module and is used for carrying out segmentation and feature extraction on the water surface spray image acquired in real time by adopting a convolutional neural network to obtain spray features of the water surface spray image, and the spray features comprise the size and the duration of the spray; the water spray features before feeding, during feeding and after feeding are stored in a feeding water spray image database; and the satiation degree evaluation module is used for evaluating the ingestion active degree and the satiation degree of the fish school according to the ingestion spray characteristics on the basis of a satiation degree evaluation model. According to the method, the feeding speed is adjusted according to the collected and analyzed water spray characteristics on the water surface, so that the feeding speed is coordinated with the feeding speed, precise feeding is achieved, and residual feed is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of aquaculture, and in particular to a precise feeding system and method based on machine vision. Background Art

[0002] my country is the world's largest aquaculture country, accounting for about 70% of the world's total aquaculture output. This huge scale of aquaculture has a profound and positive impact on my country's national economy and social development. Especially in freshwater fish farming, my country's geographical and climatic conditions provide a good breeding environment, making freshwater fish farming an important part of my country's aquaculture. This not only helps to meet the domestic residents' demand for protein, but also creates huge economic value for exports.

[0003] However, despite the large scale of aquaculture in my country, actual production faces serious challenges, such as low feeding efficiency and environmental pollution. Under the current feeding method, about 15% of the bait is not effectively eaten and settles at the bottom of the fish pond or dissolves in the water, which increases the carbon, nitrogen and phosphorus content in the aquaculture water, causing aquaculture environmental pollution and increased aquaculture risks. Accurately analyzing the feeding status of fish and accurately feeding are of great significance for reducing pollution and aquaculture costs and ensuring the survival rate of fish.

[0004] In order to better leverage the advantages of aquaculture and promote the sustainable development of the industry, it is necessary to adopt innovative methods and scientific and technological means to improve breeding efficiency, reduce environmental burden, and ensure that aquaculture plays a more active role in the national economy and society. Summary of the invention

[0005] In view of the difficulty in accurate feeding in existing fish farming, which easily produces residual bait and affects water quality, the present invention provides a precise feeding system and method based on machine vision. In the present invention, accurate feeding is achieved by real-time collection of water surface splash images and analysis of water surface splash characteristics for feeding, which can reduce the generation of residual bait; in addition, the image processing module of the present invention analyzes the splash characteristics of the three stages of fish before feeding, during feeding, and after feeding, further achieving accurate feeding and avoiding the occurrence of residual bait or incomplete satiety caused by overfeeding of fish.

[0006] The technical solution adopted by the present invention is:

[0007] A precision feeding system based on machine vision, comprising an image acquisition module, an image processing module, a feeding splash image database, a satiety assessment module, an automatic feeding machine and a control module;

[0008] Image acquisition module, used for real-time acquisition of water splash images on the fish pond;

[0009] The image processing module is connected to the image acquisition module signal, and uses a convolutional neural network to segment and extract features of the water surface splash images collected in real time, and obtains the splash features of the water surface splash images, the splash features including the size and duration of the splash; and stores the splash features before, during and after feeding into the feeding splash image database;

[0010] The satiety assessment module is based on the satiety assessment model and assesses the feeding activity and satiety of fish schools according to the characteristics of feeding splashes;

[0011] The control module is connected with the image processing module, the satiety assessment module, and the automatic feeding machine signal, and determines the feeding start time and feeding end time according to the splash characteristics of the real-time water surface splash image obtained by the image processing module and the feeding activity and satiety of the fish school obtained by the satiety assessment module, and determines the estimated feeding amount and feeding speed according to the satiety of the fish school and the interval time from the last feeding, and sends a feeding control instruction to the automatic feeding machine to control the operation of the automatic feeding machine;

[0012] The automatic feeding machine is used to put bait into the fish pond according to the control instructions of the control module.

[0013] Furthermore, the image acquisition module includes a camera for acquiring images and an LED light source for supplementing light to the fish pond.

[0014] Furthermore, before feeding is 10 to 20 minutes before the feeding time point; after feeding is 10 to 20 minutes after the automatic feeding machine stops feeding.

[0015] Furthermore, the satiety assessment model is an LSTM (long short-term memory) neural network model of feeding splash characteristics, feeding activity, and satiety based on the biological characteristics of fish, the experience of fishery experts, and the splash characteristics of the water surface during fish farming;

[0016] The construction process of the satiety assessment model is as follows:

[0017] (1) Data preprocessing: Clean the collected data to remove invalid or outliers; normalize the data to ensure that different features have equal weights in the model; convert the time series data into the shape required by the LSTM model, i.e. [samples, time steps, features];

[0018] (2) Model construction

[0019] Choose libraries, use the Python programming language, and install NumPy for mathematical operations, TensorFlow, and Keras;

[0020] Build an LSTM model, use Keras' Sequential model, add an LSTM layer, add a fully connected layer; select the activation function and loss function, and compile the model;

[0021] (3) Model training: Use the preprocessed data to train the LSTM model, specify the number of iterations and batch size; during the training process, monitor the performance of the model and prevent overfitting.

[0022] Furthermore, the activation function is ReLU or sigmoid, and the loss function is mean square error or cross entropy.

[0023] The feeding method of the precision feeding system based on machine vision comprises the following steps:

[0024] S1. The image acquisition module acquires water surface splash images in real time, and the image processing module uses a convolutional neural network to segment and extract features of the acquired water surface splash images to obtain splash features of the water surface splash images;

[0025] S2. The satiety assessment module determines the satiety of the fish school based on the splash characteristics of the water surface splash image, and the control module determines the feeding start time, estimated feeding amount, and initial feeding speed based on the satiety and feeding time interval; and controls the feeding speed and feeding end time of the automatic feeder based on the splash characteristics during feeding after the start of feeding.

[0026] Furthermore, the control module has a built-in threshold for the splash feature before feeding. When the splash feature size of the water surface splash image before feeding is greater than the set threshold and lasts for 4 to 6 minutes, the automatic feeder is controlled to start feeding; and the estimated feeding amount and initial feeding speed are determined based on the interval time from the last feeding and the satiety assessment model.

[0027] Furthermore, the water splash image features, feeding activity, satiety, actual feeding amount and feeding speed within 10 to 20 minutes before feeding, during feeding and 10 to 20 minutes after feeding is stopped are used to update and optimize the satiety assessment module.

[0028] Furthermore, when the feeding time interval is greater than 6 hours, the single feeding time is 30 minutes to 60 minutes.

[0029] Furthermore, the convolutional neural network includes a convolution layer and a connection layer, the convolution layer includes convolution, activation function, and pooling layer; the image processing module uses a unified convolution kernel to perform sliding convolution on the image; the activation function uses the rectified linear function ReLU, and the function expression is f(x)=max(0,x); the pooling layer uses maximum pooling or mean pooling; when building the convolutional neural network model, the gradient descent method and the back propagation algorithm are used, and the convolutional neural network model is first trained using some samples.

[0030] The beneficial effects of the present invention are:

[0031] In the present invention, water surface splash images are collected in real time and the characteristics of water surface splash are analyzed for feeding, so that accurate feeding is achieved and the generation of residual bait can be reduced. In addition, the image processing module of the present invention analyzes the splash characteristics of the three stages of fish before feeding, during feeding and after feeding, so as to further achieve accurate feeding and avoid the occurrence of residual bait or incomplete satiety caused by overfeeding of fish. Too much residual bait will affect the water quality and cause waste of bait, while incomplete satiety is not conducive to the growth of fish.

[0032] The present invention determines whether to feed in advance and the initial feeding speed according to the water splash characteristics before feeding, controls the feeding time of the fish school by floating, facilitates accurate feeding control, makes feeding more in line with the actual feeding state of the fish school, changes the feeding speed in real time according to the feeding state of the fish school, facilitates the control of bait feeding, realizes accurate feeding, can reduce the generation of residual bait, avoids the influence of excessive residual bait on water quality, reduces the frequency of changing water due to deterioration of water quality, and reduces the waste of bait at the same time; adjusts the feeding speed in real time according to the water splash characteristics during feeding, determines whether to feed again according to the water splash characteristics after feeding, avoids the situation that the fish school is not fully fed, and incomplete feeding is not conducive to the growth of the fish school.

[0033] In the present invention, the water splash image features, feeding activity, satiety, actual feeding amount and feeding speed within 10 to 20 minutes before feeding, during feeding and 10 to 20 minutes after feeding is stopped can also be used to update and optimize the satiety assessment module. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of the machine vision-based precision feeding system adjusting the feeding strategy according to the water splashes described in the present invention.

[0035] Figure 2 This is a schematic diagram of the structure of the precise feeding system based on machine vision according to the present invention.

[0036] Figure 3 This is a side view of the machine vision-based precision feeding system of the present invention.

[0037] In the figure, 11. first camera, 12. LED light source, 2. computer, 3. control module, 31. embedded processing unit, 4. automatic feeding machine. DETAILED DESCRIPTION

[0038] The present invention will be further described in detail below with reference to the specific implementation modes of the accompanying drawings, but the protection scope of the present invention is not limited thereto.

[0039] Figure 1-Figure 3This is a specific embodiment of the machine vision-based precision feeding system of the present invention, and the system includes an image acquisition module, an image processing module, a feeding splash image database, a satiety assessment module, an automatic feeding machine 4 and a control module 13. The automatic feeding machine 4 is arranged at the edge of the fish pond and is used to continuously feed bait into the fish pond at a certain feeding speed.

[0040] The image acquisition module is used to acquire the water splash image of the fish pond in real time. The image acquisition module includes a camera 11 for acquiring images and an LED light source 12 for supplementing light for the fish pond. The LED light source 12 can provide supplementary light at night or on cloudy days or other conditions with poor light, so that the camera 11 can acquire the water surface image.

[0041] The image processing module is connected to the image acquisition module signal, and the image processing module can use computer 2 to process pictures. The image processing module uses a convolutional neural network to segment and extract features of the water surface image transmitted by the image acquisition module, and analyze the water splash features, and the analyzed water splash features include the size and duration of the water splash. The image processing module continuously analyzes the water surface splash features of the fish before, during and after feeding, and stores the water splash features before, during and after feeding into the feeding water splash image database. Among them, before feeding is 10 to 20 minutes before the feeding time point, and after feeding is 10 to 20 minutes after the automatic feeding machine 4 stops feeding.

[0042] Convolutional neural network includes convolution layer and connection layer. Convolution layer includes convolution, activation function and pooling layer. Among them, a uniform convolution kernel is used to perform sliding convolution on the image, so that the weight parameters do not change, which also reduces the number of parameters to be trained. For convolutional neural network, the parameters are the elements of the convolution kernel. When the uniform convolution is used to perform convolution on the whole image, the parameter sharing in the convolution process is realized. The combination of parameter sharing and pooling layer enables convolutional neural network to effectively process image data and is robust to the position change of objects in the image. The parameter sharing process makes it possible to extract features from the output even if the input object moves, although there may be differences in the position of the extracted features. For example, for an original image, a convolution kernel for extracting edges is used. Even if the image in the original image is translated, the features in the image after the convolution operation are only partially displaced, but image extraction can still be achieved.

[0043] The activation function in the second level uses the rectified linear function ReLU, and the function expression is f(x) = max(0,x). Compared with the Tanh function, the ReLU function is not prone to the problem of gradient saturation during the gradient back propagation process, that is, when the neuron input is too large or too small, the gradient is close to 0. The reason for this phenomenon is that the derivative of the activation function will be multiplied during the gradient back propagation process. Obviously, when the input is too large, the derivative tends to 0, and when the gradient is saturated, the result of the gradient descent method will converge very slowly. In addition, the ReLU function is a nonlinear function. The use of nonlinear functions as activation functions has a great influence on the network. If the activation functions in the second layer are all linear, the entire network is also linear, which means that the network cannot solve the corresponding linear inseparable problem, limiting the network's expression ability. For example, a neural network with a single hidden layer containing a linear activation function cannot solve the XOR problem, but a nonlinear activation function such as the ReLU function can solve it.

[0044] The pooling layer in the third level uses maximum pooling or mean pooling to reduce the dimension of the feature map, reduce the number of parameters in the network, avoid overfitting, and improve the calculation speed and operation efficiency of the model. The pooling layer outputs the pooled features to the connection layer.

[0045] In convolutional neural networks, the connection layer uses sparse interaction to reduce the number of calculation parameters and calculation steps. For example, when the two layers of the neural network are fully connected, assuming that the number of neurons in the two layers of the neural network is M, M×N parameters are required; when the sparse connection method is used, not every neuron is connected to all neurons in the next layer, but only to a part of the neurons. Assuming that each neuron is only connected to K neurons in the next layer, in general, the parameter K is much smaller than M, then the number of required parameters is K×N, and the number of parameters will be greatly reduced. In convolutional neural networks, sparse interaction is achieved by controlling the size of the convolution kernel, and the size of the convolution kernel determines the degree of sparse interaction.

[0046] When building a convolutional neural network model, it is necessary to first use some samples to train the convolutional neural network model. In this embodiment, the convolutional neural network model is trained using the gradient descent method and the back propagation algorithm.

[0047] The satiety assessment module is based on the satiety assessment model and assesses the feeding activity and satiety of fish according to the feeding splash characteristics. The satiety assessment model is an LSTM (long short-term memory) neural network model of feeding splash characteristics, feeding activity, and satiety constructed based on the biological characteristics of fish, the experience of fishery experts, and the splash characteristics of the water surface during fish breeding.

[0048] The construction process of the satiety assessment model is as follows:

[0049] (1) Data preprocessing: Clean the collected data to remove invalid or outliers; normalize the data to ensure that different features have equal weights in the model; and convert the time series data into the shape required by the LSTM model, i.e., [samples, time steps, features].

[0050] (2) Model construction

[0051] Choose libraries, use the Python programming language, and install NumPy for mathematical operations, TensorFlow, and Keras;

[0052] Build an LSTM model, use the Sequential model of Keras, add an LSTM layer, add a fully connected layer, select an activation function and a loss function, and compile the model. The activation function is ReLU or sigmoid, and the loss function is mean square error or cross entropy.

[0053] (3) Model training: Use the preprocessed data to train the LSTM model, specify the number of iterations and batch size; during the training process, monitor the performance of the model and prevent overfitting.

[0054] The control module 13 is connected with the image processing module, the satiety assessment module, and the automatic feeder 4 by signal. According to the splash characteristics of the real-time water splash image obtained by the image processing module, the feeding activity and satiety of the fish school obtained by the satiety assessment module, the feeding start time, the estimated feeding amount, the feeding speed, and the feeding end time are determined, and a feeding control instruction is sent to the automatic feeder 4 to control the operation of the automatic feeder 4. Specifically, the image acquisition module collects the surface splash image of the fish pond in real time, and obtains the splash characteristics of the surface splash image through the image processing module. When the satiety assessment module evaluates that the splash characteristics at this moment meet the splash characteristics before feeding according to the acquired splash characteristics, and lasts for 4 to 6 minutes, the control module 13 controls the automatic feeder to start feeding, otherwise, no advance feeding is performed. The estimated feeding amount and initial feeding speed are determined based on the interval time from the last feeding and the satiety assessment model, that is, the estimated feeding amount and initial feeding speed are determined based on the analyzed water splash characteristics before feeding and the feeding activity of the fish school obtained by the satiety assessment module. For example, if the feeding time is set at 8:00, and the water splash image captured before the feeding time meets the feeding start conditions, the feeding will be started in advance; if the water splash image captured at 8:00 still does not meet the feeding start conditions, the feeding time will be postponed. When the water splash characteristics reach the water splash characteristics after feeding, the feeding is stopped. By controlling the feeding time of the fish school by floating, it is convenient to accurately control the feeding, so that the feeding is more in line with the actual feeding status of the fish school. The time interval from the last feeding can reflect the satiety of the fish to a certain extent. The control module 13 determines the estimated feeding amount and the initial feeding speed according to the feeding activity and satiety of the fish obtained by the satiety assessment module, and evaluates the feeding status of the fish in real time according to the changes in the water splashes during the facility process, and adjusts the feeding speed in real time, which is convenient for controlling the feeding, achieving accurate feeding, and reducing the generation of residual bait.

[0055] The control module 13 determines whether to feed again according to the splash characteristics after feeding analyzed by the image processing module; when the splash characteristics after feeding obtained by the image processing module are judged based on the satiety evaluation module that the fish still has the desire to eat, and the duration reaches 4 to 6 minutes, the control module 13 controls the automatic feeding machine 4 to feed again according to the corresponding feeding speed. Because the bait sinks, the fish dives into the bottom of the water to feed, and the camera cannot capture the splash image, and mistakenly thinks that the feeding is over. If after a period of time, it is found by capturing the splash image that the fish floats again and still has the desire to eat, and feeds again to avoid the situation where the fish are not fully fed; in addition, the water splash duration is set to 4 to 6 minutes to avoid the situation of misjudgment. When the feeding time interval is greater than 6 hours, the single feeding duration of the automatic feeding machine 4 when feeding bait into the fish pond is 30min-60min, and the single feeding duration corresponds to the feeding duration of the fish, which is convenient for observing the feeding state of the fish.

[0056] In this embodiment, the control module 3 uses an embedded processing unit 31. In addition, the splash image features, feeding activity, satiety, and actual feeding amount and feeding speed within 10 to 20 minutes before feeding, during feeding, and 10 to 20 minutes after feeding is stopped are used to update and optimize the satiety assessment module.

[0057] The feeding method according to the precision feeding system based on machine vision comprises the following steps:

[0058] S1. The image acquisition module acquires water surface splash images in real time, and the image processing module uses a convolutional neural network to segment and extract features of the acquired water surface splash images to obtain the splash features of the water surface splash images.

[0059] S2. The satiety assessment module determines the satiety of the fish school based on the splash characteristics of the water splash image on the water surface. The control module 13 determines the feeding start time, estimated feeding amount, and initial feeding speed based on the satiety and feeding time interval; and controls the feeding speed and feeding end time of the automatic feeder 4 based on the splash characteristics during feeding after the start of feeding.

[0060] The control module has a built-in threshold for the splash feature before feeding. When the splash feature size of the water surface splash image before feeding is greater than the set threshold and lasts for 4 to 6 minutes, the automatic feeder is controlled to start feeding; and the estimated feeding amount and initial feeding speed are determined based on the interval time from the last feeding and the satiety assessment model. The splash image features, feeding activity, satiety, and actual feeding amount and feeding speed within 10 to 20 minutes before feeding, during feeding, and 10 to 20 minutes after feeding are stopped are used to update and optimize the satiety assessment module. When the feeding time interval is greater than 6 hours, the single feeding time is 30 minutes to 60 minutes.

[0061] In the present invention, the water splash image of the water surface is collected in real time and the water splash characteristics are analyzed for feeding, so as to realize accurate feeding and reduce the generation of residual bait; in addition, the image processing module of the present invention analyzes the water splash characteristics of the three stages before, during and after the feeding of the fish school, so as to further realize accurate feeding and avoid the situation that the fish school is not fully fed or overfed, resulting in residual bait. Whether to feed in advance and the initial feeding speed are determined according to the water splash characteristics before feeding, and the feeding time of the fish school is controlled by floating, so as to facilitate accurate feeding control, so that the feeding is more in line with the actual feeding state of the fish school, and the feeding speed is changed in real time according to the feeding state of the fish school, so as to facilitate the control of feeding, realize accurate feeding, reduce the generation of residual bait, avoid excessive residual bait affecting the water quality, and reduce the frequency of changing water due to deterioration of water quality; at the same time, reduce the waste of bait; adjust the feeding speed in real time according to the water splash characteristics during feeding, and determine whether to feed again according to the water splash characteristics after feeding, so as to avoid the situation that the fish school is not fully fed, and incomplete feeding is not conducive to the growth of the fish school. In addition, the water splash image features, feeding activity, satiety level, actual feeding amount and feeding speed within 10 to 20 minutes before feeding, during feeding and 10 to 20 minutes after feeding is stopped can also be used to update and optimize the satiety assessment module.

[0062] The examples are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A precision feeding system based on machine vision, characterized in that: It includes an image acquisition module, an image processing module, a water splash image database, a satiety assessment module, an automatic feeding machine (4) and a control module (13); Image acquisition module, used for real-time acquisition of water splash images on the fish pond; The image processing module is connected to the image acquisition module by signal, and uses a convolutional neural network to segment and extract features of the water surface splash image acquired in real time, so as to obtain the splash features of the water surface splash image, wherein the splash features include the size and duration of the splash; and storing the water splash features before, during and after feeding into a feeding water splash image database; The satiety assessment module assesses the feeding activity and satiety of fish schools based on the satiety assessment model and the characteristics of feeding splashes; The control module (13) is connected to the image processing module, the satiety evaluation module, and the automatic feeding machine (4) by signals, determines the feeding start time and feeding end time based on the water splash characteristics of the real-time water surface splash image obtained by the image processing module and the feeding activity and satiety of the fish school obtained by the satiety evaluation module, determines the estimated feeding amount and feeding speed based on the satiety of the fish school and the interval time from the last feeding, and sends a feeding control instruction to the automatic feeding machine (4) to control the operation of the automatic feeding machine (4); The automatic feeding machine (4) is used to put bait into the fish pond according to the control instructions of the control module.

2. The machine vision-based precision feeding system according to claim 1, characterized in that: The image acquisition module comprises a camera (11) for acquiring images, and an LED light source (12) for supplementing light to the fish pond.

3. The machine vision-based precision feeding system according to claim 1, characterized in that: Before feeding is 10 to 20 minutes before the feeding time point; after feeding is 10 to 20 minutes after the automatic feeding machine (4) stops feeding.

4. The machine vision-based precision feeding system according to claim 1, characterized in that: The satiety assessment model is an LSTM (long short-term memory) neural network model of feeding splash characteristics, feeding activity, and satiety based on the biological characteristics of fish, the experience of fishery experts, and the splash characteristics of the water surface during fish farming. The construction process of the satiety assessment model is as follows: (1) Data preprocessing: Clean the collected data to remove invalid or outliers; normalize the data to ensure that different features have equal weights in the model; convert the time series data into the shape required by the LSTM model, i.e. [samples, time steps, features]; (2) Model construction Choose libraries, use the Python programming language, and install NumPy for mathematical operations, TensorFlow, and Keras; Build an LSTM model, use Keras' Sequential model, add an LSTM layer, add a fully connected layer; select the activation function and loss function, and compile the model; (3) Model training: Use the preprocessed data to train the LSTM model, specify the number of iterations and batch size; during the training process, monitor the performance of the model and prevent overfitting.

5. The machine vision-based precision feeding system according to claim 4, characterized in that: The activation function is ReLU or sigmoid, and the loss function is mean square error or cross entropy.

6. The feeding method of the precision feeding system based on machine vision according to any one of claims 1 to 5, characterized in that: The following steps are involved: S1. The image acquisition module acquires water surface splash images in real time, and the image processing module uses a convolutional neural network to segment and extract features of the acquired water surface splash images to obtain splash features of the water surface splash images; S2. The satiety assessment module determines the satiety of the fish school based on the splash characteristics of the water splash image on the water surface. The control module (13) determines the feeding start time, estimated feeding amount, and initial feeding speed based on the satiety and feeding time interval; and controls the feeding speed and feeding end time of the automatic feeder (4) based on the splash characteristics during feeding after the start of feeding.

7. The feeding method according to claim 6, characterized in that: The control module has a built-in threshold for the splash feature before feeding. When the splash feature size of the water surface splash image before feeding is greater than the set threshold and lasts for 4 to 6 minutes, the automatic feeder is controlled to start feeding; and the estimated feeding amount and initial feeding speed are determined based on the interval time from the last feeding and the satiety evaluation model.

8. The feeding method according to claim 6, characterized in that: The water splash image features, feeding activity, satiety, actual feeding amount and feeding speed within 10 to 20 minutes before feeding, during feeding and 10 to 20 minutes after feeding is stopped are used to update and optimize the satiety assessment module.

9. The feeding method according to claim 6, characterized in that: When the feeding interval is greater than 6 hours, the single feeding time is 30 minutes to 60 minutes.

10. The feeding method according to claim 6, characterized in that: The convolutional neural network includes a convolution layer and a connection layer, and the convolution layer includes a convolution, an activation function, and a pooling layer; the image processing module uses a unified convolution kernel to perform sliding convolution on the image; the activation function uses a rectified linear function ReLU, and the function expression is f(x)=max(0,x); the pooling layer uses maximum pooling or mean pooling; when building a convolutional neural network model, the gradient descent method and the back propagation algorithm are used, and the convolutional neural network model is first trained using some samples.

Images