Method for estimating fry quantity based on multi-scale semantic fusion density estimation network

Abstract

The present invention relates to the field of image processing technology, specifically a method for estimating the number of fry based on a multi-scale semantic fusion density estimation network, including: The present invention first divides the fish pond into grids, and obtains the first fry image to be detected, the light intensity and the area of each grid region; constructs a fry number estimation model to process the image to be detected, obtains the gray-level co-occurrence matrix of the fry image to be detected through image recognition, obtains the texture features, and further obtains the first turbidity coefficient, and determines different denoising methods according to the first turbidity coefficient; the feature extraction layer is processed by a large-scale feature extraction unit, a small-scale feature extraction unit and an attention mechanism; the feature fusion layer obtains the first fry fusion feature map; the feature analysis layer uses the first fry fusion feature map to generate the first fry density map, obtains the first fry number, and corrects it according to the light intensity to obtain the second fry number; this method can effectively improve the accuracy of fry number estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and specifically to a method for estimating the number of fry based on a multi-scale semantic fusion density estimation network. Background Art

[0002] During the aquaculture process, estimating the number of fry is an important part of production management. Currently, most traditional methods for estimating the number of fry estimate the fish quantity through single-scale feature analysis and do not fully consider the impact of environmental factors on fry quantity estimation, which easily leads to limited accuracy in estimating the number of fry.

[0003] Therefore, a method for estimating the number of fry based on a multi-scale semantic fusion density estimation network is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for estimating the number of fry based on a multi-scale semantic fusion density estimation network. The present invention relates to the technical field of image processing, and specifically to a method for estimating the number of fry based on a multi-scale semantic fusion density estimation network, including: The present invention first divides the fish pond into grids to obtain the first fry image to be detected, the light intensity, and the area of each grid area in each grid region; constructs a fry number estimation model to process the image to be detected, obtains the gray-level co-occurrence matrix of the fry image to be detected through image recognition to obtain texture features, and further obtains a first turbidity coefficient, and determines different denoising methods according to the first turbidity coefficient; the feature extraction layer is processed through a large-scale feature extraction unit, a small-scale feature extraction unit, and an attention mechanism; the feature fusion layer obtains a first fry fusion feature map; the feature analysis layer uses the first fry fusion feature map to generate a first fry density map to obtain a first fry number, and corrects it according to the light intensity to obtain a second fry number; this method can effectively improve the accuracy of fry number estimation.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for estimating the number of fry based on a multi-scale semantic fusion density estimation network, including:

[0007] S1. Divide the fish pond into grids to obtain each fish pond grid region, and obtain the first fry image to be detected, the second light intensity, and the third area of each fish pond grid region;

[0008] S2. Construct a fry quantity estimation model to identify the first fry image to be detected. The fry quantity estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; the preprocessing layer performs image recognition on the first fry image to be detected to obtain a first turbidity coefficient; determines different denoising methods according to the first turbidity coefficient and a preset turbidity coefficient threshold set; and performs multi-scale image segmentation on the first fry image to be detected to obtain multiple first large-scale fry detection images and second small-scale fry detection images;

[0009] S3. The feature extraction layer respectively extracts features from the first large-scale fry detection image and the second small-scale fry detection image through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes them through an SE attention mechanism to respectively obtain a first large-scale feature map and a second small-scale feature map; the feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain a first fry fusion feature map;

[0010] S4. The feature analysis layer obtains a first fry density map according to the first fry fusion feature map; obtains a first fry number according to the first fry density map and a third area; and obtains a second fry number according to the first fry number, a second light intensity, and the first turbidity coefficient.

[0011] Preferably, the preprocessing layer performs image recognition on the first fry image to be detected to obtain the first turbidity coefficient specifically including:

[0012] S10. Grayscale the first fry image to be detected and calculate the gray-level co-occurrence matrix of the first fry image to be detected;

[0013] S20. Extract texture features according to the gray-level co-occurrence matrix, and the texture features include contrast, uniformity, and energy;

[0014] S30. Obtain the first turbidity coefficient according to the texture features.

[0015] Preferably, the determining different denoising methods according to the first turbidity coefficient and a preset turbidity coefficient threshold set specifically includes:

[0016] S100. Determine the turbidity level of the fish pond grid area according to the first turbidity coefficient and a preset turbidity coefficient threshold set, and the turbidity level includes mild turbidity, moderate turbidity, and severe turbidity;

[0017] S200. Denoise the mildly turbid fish pond grid area by Gaussian filtering; denoise the moderately turbid fish pond grid area by bilateral filtering; denoise the severely turbid fish pond grid area by deep learning denoising.

[0018] Preferably, the preset turbidity coefficient threshold set includes a first turbidity coefficient threshold and a second turbidity coefficient threshold; the specific process for obtaining the turbidity level is as follows:

[0019] ;

[0020] wherein, represents the turbidity level; represents mild turbidity; represents the first turbidity coefficient; represents the first turbidity coefficient threshold; represents the second turbidity coefficient threshold; represents moderate turbidity; represents severe turbidity.

[0021] Preferably, the fry quantity estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer;

[0022] The preprocessing layer performs image recognition on the first fry image to be detected to obtain the first turbidity coefficient, and determines different denoising methods according to the first turbidity coefficient; and performs multi-scale image segmentation on the first fry image to be detected through U-Net to obtain a plurality of first large-scale fry detection images and second small-scale fry detection images;

[0023] The feature extraction layer respectively extracts features from the first large-scale fry detection image and the second small-scale fry detection image through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes them through an SE attention mechanism to obtain a first large-scale feature map and a second small-scale feature map respectively;

[0024] The feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain a first fry fusion feature map;

[0025] The feature analysis layer reduces the number of channels of the first fry fusion feature map through a first convolutional kernel, and sequentially passes through a second convolutional kernel and a third convolutional kernel to obtain a first fry density map; according to the first fry density map and the third area, the first fry quantity is obtained; and according to the first fry quantity, the second light intensity, and the first turbidity coefficient, the second fry quantity is obtained;

[0026] The output layer is used to output the second fry quantity.

[0027] Preferably, the first fry quantity is:

[0028] ;

[0029] wherein, represents the first fry quantity; represents the third area; represents the number of fishpond grid areas; represents the total number of pixels of the first fry density map; represents the sum of the pixel values of the first fry density map.

[0030] Preferably, the number of the second fry is:

[0031] ;

[0032] wherein, represents the number of the second fry; represents the number of the first fry; represents the weight of the influence of light intensity; represents the average light intensity of the fishpond; represents the reference value of light intensity.

[0033] Preferably, the average light intensity of the fishpond is obtained according to the second light intensity and the number of fishpond grid areas.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. The present invention obtains the texture features of the fry image to be detected by recognizing the gray-level co-occurrence matrix of the fry image to be detected, further obtains the first turbidity coefficient according to the texture features, determines the turbidity level of the fishpond grid area according to the first turbidity coefficient and the preset turbidity coefficient threshold, and determines different denoising methods according to the turbidity level; through this method, the influence of fishpond turbidity on the estimation of fry quantity can be effectively avoided, providing a good data image basis for the later model analysis, and further effectively improving the accuracy of fry quantity estimation.

[0036] 2. The present invention uses a fry quantity estimation model to recognize the fry image to be detected, divides the fry image to be detected in each fishpond grid area into large-scale images and small-scale images through image segmentation, the feature extraction layer distributes to extract features through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes through the SE attention mechanism to obtain a first large-scale feature map and a second small-scale feature map; then, according to the feature fusion layer, the first large-scale feature map and the second small-scale feature map are fused to obtain a first fry fusion feature map; according to the first fry fusion feature map, a first fry density map is obtained, and then the number of the first fry is obtained. This method can effectively integrate the multi-scale features of the fry image to be detected through the fry quantity estimation model, effectively improve the recognition ability of the fry image to be detected, and further effectively improve the accuracy of fry quantity estimation.

[0037] 3. The present invention analyzes by combining the light intensity and the number of first fry in each fish pond grid area, and further obtains the number of second fry. Through effective correction, the influence of light intensity on the estimation of the number of fry can be avoided, and the accuracy of the estimation of the number of fry can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic flow chart of the fry number estimation method based on the multi-scale semantic fusion density estimation network provided by the embodiment of the present invention;

[0039] Figure 2 Schematic diagram of a denoising process provided by the embodiment of the present invention;

[0040] Figure 3 Schematic diagram of the structure of the fry number estimation model provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0042] Embodiment 1

[0043] In order to improve the accuracy of estimating the number of fry in Pond A, the fry number estimation method based on the multi-scale semantic fusion density estimation network is applied;

[0044] Refer to Figure 1 The schematic flow chart of the fry number estimation method based on the multi-scale semantic fusion density estimation network provided by the embodiment of the present invention, specifically including:

[0045] S1. Divide the fish pond into grid areas to obtain each fish pond grid area, and obtain the first to-be-detected fry image, the second light intensity, and the third area of each fish pond grid area;

[0046] Further, in this embodiment, the number of fish pond grid areas is N; the third area represents the area of each fish pond grid area;

[0047] S2. Construct a fry quantity estimation model to identify the first fry image to be detected. The fry quantity estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; the preprocessing layer performs image recognition on the first fry image to be detected to obtain a first turbidity coefficient; different denoising methods are determined according to the first turbidity coefficient and a preset set of turbidity coefficient thresholds; and multi-scale image segmentation is performed on the first fry image to be detected to obtain multiple first large-scale fry detection images and second small-scale fry detection images;

[0048] Further, the fry quantity estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; as Figure 3 is a schematic structural diagram of a fry quantity estimation model provided by an embodiment of the present invention;

[0049] The preprocessing layer performs image recognition on the first fry image to be detected to obtain a first turbidity coefficient, and determines different denoising methods according to the first turbidity coefficient; and performs multi-scale image segmentation on the first fry image to be detected through U-Net to obtain multiple first large-scale fry detection images and second small-scale fry detection images;

[0050] The feature extraction layer respectively extracts features of the first large-scale fry detection image and the second small-scale fry detection image through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes them through an SE attention mechanism to respectively obtain a first large-scale feature map and a second small-scale feature map;

[0051] The feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain a first fry fusion feature map;

[0052] The feature analysis layer reduces the number of channels of the first fry fusion feature map through a first convolutional kernel, and sequentially passes through a second convolutional kernel and a third convolutional kernel to obtain a first fry density map; the size of the first convolutional kernel is 1×1; the size of the second convolutional kernel is 3×3; the size of the third convolutional kernel is 1×1; according to the first fry density map and a third area, a first fry number is obtained; and a second fry number is obtained according to the first fry number, a second light intensity, and the first turbidity coefficient;

[0053] The output layer is used to output the second fry number.

[0054] Further, the preprocessing layer performing image recognition on the first fry image to be detected to obtain a first turbidity coefficient specifically includes:

[0055] S10. Grayscale the first fry image to be detected, and calculate the gray-level co-occurrence matrix of the first fry image to be detected;

[0056] S20. Extract texture features according to the gray-level co-occurrence matrix, where the texture features include contrast, uniformity, and energy;

[0057] S30. Obtain the first turbidity coefficient according to the texture features;

[0058] Further, the first turbidity coefficient is:

[0059] ;

[0060] where, represents the first turbidity coefficient; represents the contrast influence weight; represents the contrast, reflecting the gray-scale difference; represents the uniformity influence weight; represents the uniformity, reflecting the texture regularity; represents the energy influence weight; represents the energy, and the energy reflects the texture smoothness;

[0061] Further, the determining of different denoising methods according to the first turbidity coefficient and the preset turbidity coefficient threshold set specifically includes:

[0062] S100. Determine the turbidity level of the fishpond grid area according to the first turbidity coefficient and the preset turbidity coefficient threshold set, where the turbidity level includes mild turbidity, moderate turbidity, and severe turbidity;

[0063] S200. Denoise the mildly turbid fishpond grid area through Gaussian filtering; denoise the moderately turbid fishpond grid area through bilateral filtering; denoise the severely turbid fishpond grid area through deep learning denoising.

[0064] Further, the preset turbidity coefficient threshold set includes a first turbidity coefficient threshold and a second turbidity coefficient threshold; the specific obtaining process of the turbidity level is:

[0065] ;

[0066] where, represents the turbidity level; represents mild turbidity; represents the first turbidity coefficient; represents the first turbidity coefficient threshold; represents the second turbidity coefficient threshold; represents moderate turbidity; represents severe turbidity;

[0067] The first turbidity coefficient threshold and the second turbidity coefficient threshold are obtained through training with historical fishpond image data;

[0068] The schematic diagram of the specific denoising process is as follows Figure 2 shown;

[0069] In this embodiment, the gray-level co-occurrence matrix of the fry image to be detected is recognized through image recognition, and the texture features of the fry image to be detected are further obtained. The first turbidity coefficient is obtained according to the texture features, and the turbidity level of the fish pond grid area is determined according to the first turbidity coefficient and the preset turbidity coefficient threshold, and different denoising methods are determined according to the turbidity level. By this method, the influence of fish pond turbidity on the estimation of fry quantity can be effectively avoided, providing a good data image basis for the later model analysis and further effectively improving the accuracy of fry quantity estimation.

[0070] S3. The feature extraction layer respectively extracts the features of the first large-scale fry detection image and the second small-scale fry detection image through the large-scale feature extraction unit and the small-scale feature extraction unit, and processes them through the SE attention mechanism to obtain the first large-scale feature map and the second small-scale feature map respectively; the feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain the first fry fusion feature map;

[0071] S4. The feature analysis layer obtains the first fry density map according to the first fry fusion feature map; obtains the first fry number according to the first fry density map and the third area; and obtains the second fry number according to the first fry number, the second light intensity and the first turbidity coefficient;

[0072] Furthermore, the first fry number is:

[0073] ;

[0074] Wherein, represents the first fry number; represents the third area; represents the number of fish pond grid areas; represents the total number of pixels of the first fry density map; represents the sum of pixel values of the first fry density map.

[0075] In this embodiment, the fish fry quantity estimation model is used to identify the fish fry image to be detected. The fish fry image to be detected in each fish pond grid area is segmented into large-scale images and small-scale images through image segmentation. The feature extraction layer distributes to extract features through the large-scale feature extraction unit and the small-scale feature extraction unit, and processes them through the SE attention mechanism to obtain the first large-scale feature map and the second small-scale feature map. Furthermore, the first large-scale feature map and the second small-scale feature map are fused according to the feature fusion layer to obtain the first fish fry fusion feature map. The first fish fry density map is obtained according to the first fish fry fusion feature map, and then the first fish fry quantity is obtained. This method can effectively integrate the multi-scale features of the fish fry image to be detected through the fish fry quantity estimation model, effectively improve the recognition ability of the fish fry image to be detected, and further improve the accuracy of fish fry quantity estimation.

[0076] Furthermore, the second fish fry quantity is:

[0077] ;

[0078] Wherein, represents the second fish fry quantity; represents the first fish fry quantity; represents the light intensity influence weight; represents the average light intensity of the fish pond; represents the light intensity reference value.

[0079] Furthermore, the average light intensity of the fish pond is obtained according to the second light intensity and the number of fish pond grid areas;

[0080] The average light intensity of the fish pond is:

[0081] ;

[0082] Wherein, represents the average light intensity of the fish pond; represents the number of fish pond grid areas; represents the second light intensity;

[0083] In this embodiment, by combining the light intensity of each fish pond grid area and the first fish fry quantity for analysis, the second fish fry quantity is further obtained. By effective correction, the influence of light intensity on fish fry quantity estimation can be avoided, and the accuracy of fish fry quantity estimation is further improved.

[0084] In this embodiment, the fishpond is first divided into grids to obtain the first fish fry images to be detected, the light intensity, and the area of each grid region in the fishpond; a fish fry quantity estimation model is constructed to process the images to be detected. Through image recognition, the gray-level co-occurrence matrix of the fish fry images to be detected is obtained, and texture features are obtained. Furthermore, a first turbidity coefficient is obtained, and different denoising methods are determined according to the first turbidity coefficient; the feature extraction layer is processed by a large-scale feature extraction unit, a small-scale feature extraction unit, and an attention mechanism; the feature fusion layer obtains the first fish fry fusion feature map; the feature analysis layer uses the first fish fry fusion feature map to generate the first fish fry density map, obtains the first number of fish fry, and corrects it according to the light intensity to obtain the second number of fish fry; this method can effectively improve the accuracy of fish fry quantity estimation.

[0085] To verify the effectiveness of the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by the present invention, the accuracy of fish fry quantity estimation of different methods is compared, including Method 1, Method 2, and Method 3; Method 1 is the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by this embodiment; Method 2 does not consider denoising on the basis of Method 1; Method 3 does not consider the influence of light intensity on the basis of Method 1. The average accuracy of fish fry quantity estimation for Pond A by different methods is given, and the specific results are shown in Table 1.

[0086] Table 1 Average accuracy of fish fry quantity estimation for Pond A by different methods

[0087] Method Estimated mean accuracy Method 1 96% Method 2 89% Method 3 86%

[0088] As can be seen from Table 1, the effectiveness of the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by this embodiment can effectively improve the accuracy of fish fry quantity estimation.

[0089] Embodiment 2

[0090] To improve the accuracy of fish fry quantity estimation in Pond B, the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network is applied.

[0091] Refer to Figure 1 which is a schematic flow diagram of the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by the embodiment of the present invention, and specifically includes:

[0092] S1. Divide the fishpond into grids to obtain each fishpond grid region, and obtain the first fish fry images to be detected, the second light intensity, and the third area of each fishpond grid region; further, in this embodiment, the number of fishpond grid regions is N; the third area represents the area of each fishpond grid region.

[0093] S2. Construct a fry quantity estimation model to identify the first fry image to be detected. The fry quantity estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; the preprocessing layer performs image recognition on the first fry image to be detected to obtain a first turbidity coefficient; determines different denoising methods according to the first turbidity coefficient and a preset set of turbidity coefficient thresholds; and performs multi-scale image segmentation on the first fry image to be detected to obtain a plurality of first large-scale fry detection images and second small-scale fry detection images;

[0094] Further, the fry quantity estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; as Figure 3 is a schematic structural diagram of a fry quantity estimation model provided by an embodiment of the present invention;

[0095] The preprocessing layer performs image recognition on the first fry image to be detected to obtain a first turbidity coefficient, and determines different denoising methods according to the first turbidity coefficient; and performs multi-scale image segmentation on the first fry image to be detected through U-Net to obtain a plurality of first large-scale fry detection images and second small-scale fry detection images;

[0096] The feature extraction layer respectively extracts features from the first large-scale fry detection image and the second small-scale fry detection image through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes them through an SE attention mechanism to respectively obtain a first large-scale feature map and a second small-scale feature map;

[0097] The feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain a first fry fusion feature map;

[0098] The feature analysis layer performs channel dimensionality reduction on the first fry fusion feature map through a first convolutional kernel, and sequentially passes through a second convolutional kernel and a third convolutional kernel to obtain a first fry density map; the size of the first convolutional kernel is 1×1; the size of the second convolutional kernel is 3×3; the size of the third convolutional kernel is 1×1; obtain a first fry number according to the first fry density map and the third area; and obtain a second fry number according to the first fry number, the second light intensity, and the first turbidity coefficient;

[0099] The output layer is used to output the second fry number.

[0100] Further, the preprocessing layer performing image recognition on the first fry image to be detected to obtain a first turbidity coefficient specifically includes:

[0101] S10. Grayscale the first fry image to be detected, and calculate the gray-level co-occurrence matrix of the first fry image to be detected;

[0102] S20. Extract texture features according to the gray-level co-occurrence matrix, where the texture features include contrast, uniformity, and energy;

[0103] S30. Obtain the first turbidity coefficient according to the texture features;

[0104] Further, the first turbidity coefficient is:

[0105] ;

[0106] where represents the first turbidity coefficient; represents the contrast influence weight; represents the contrast, reflecting the gray-level difference; represents the uniformity influence weight; represents the uniformity, reflecting the texture regularity; represents the energy influence weight; represents the energy, and the energy reflects the texture smoothness;

[0107] Further, the specific process of determining different denoising methods according to the first turbidity coefficient and the preset turbidity coefficient threshold set includes:

[0108] S100. Determine the turbidity level of the fishpond grid area according to the first turbidity coefficient and the preset turbidity coefficient threshold set, where the turbidity level includes mild turbidity, moderate turbidity, and severe turbidity;

[0109] S200. Denoise the fishpond grid area with mild turbidity through Gaussian filtering; denoise the fishpond grid area with moderate turbidity through bilateral filtering; denoise the fishpond grid area with severe turbidity through deep learning denoising.

[0110] Further, the preset turbidity coefficient threshold set includes a first turbidity coefficient threshold and a second turbidity coefficient threshold; the specific acquisition process of the turbidity level is:

[0111] ;

[0112] where represents the turbidity level; represents mild turbidity; represents the first turbidity coefficient; represents the first turbidity coefficient threshold; represents the second turbidity coefficient threshold; represents moderate turbidity; represents severe turbidity;

[0113] The first turbidity coefficient threshold and the second turbidity coefficient threshold are obtained through training with historical fishpond image data;

[0114] The schematic diagram of the specific denoising process is as follows Figure 2 shown;

[0115] In this embodiment, the gray-level co-occurrence matrix of the image to be detected of fry is recognized through an image, and the texture features of the image to be detected of fry are further obtained. The first turbidity coefficient is obtained according to the texture features, and the turbidity level of the fishpond grid area is determined according to the first turbidity coefficient and the preset turbidity coefficient threshold, and different denoising methods are determined according to the turbidity level; through this method, the influence of fishpond turbidity on the estimation of the number of fry can be effectively avoided, providing a good data image basis for the later model analysis, and further effectively improving the accuracy of fry number estimation.

[0116] S3. The feature extraction layer respectively extracts the features of the first large-scale fry detection image and the second small-scale fry detection image through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes them through an SE attention mechanism to obtain a first large-scale feature map and a second small-scale feature map respectively; the feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain a first fry fusion feature map;

[0117] S4. The feature analysis layer obtains a first fry density map according to the first fry fusion feature map; according to the first fry density map and the third area, the first fry number is obtained; and according to the first fry number, the second light intensity and the first turbidity coefficient, the second fry number is obtained;

[0118] Further, the first fry number is:

[0119] ;

[0120] Among them, represents the first fry number; represents the third area; represents the number of fishpond grid areas; represents the total number of pixels of the first fry density map; represents the sum of pixel values of the first fry density map.

[0121] In this embodiment, the fish fry quantity estimation model is used to identify the fish fry image to be detected. The fish fry image to be detected in each fish pond grid area is segmented into large-scale images and small-scale images through image segmentation. The feature extraction layer distributes feature extraction through a large-scale feature extraction unit and a small-scale feature extraction unit, and is processed by the SE attention mechanism to obtain a first large-scale feature map and a second small-scale feature map. Furthermore, the first large-scale feature map and the second small-scale feature map are fused according to the feature fusion layer to obtain a first fish fry fusion feature map. According to the first fish fry fusion feature map, a first fish fry density map is obtained, and then a first fish fry quantity is obtained. This method can effectively integrate the multi-scale features of the fish fry image to be detected through the fish fry quantity estimation model, effectively improve the recognition ability of the fish fry image to be detected, and further improve the accuracy of fish fry quantity estimation.

[0122] Further, the second fish fry quantity is:

[0123] ;

[0124] Wherein, represents the second fish fry quantity; represents the first fish fry quantity; represents the weight affected by the light intensity; represents the average light intensity of the fish pond; represents the reference value of the light intensity.

[0125] Further, the average light intensity of the fish pond is obtained according to the second light intensity and the number of fish pond grid areas;

[0126] The average light intensity of the fish pond is:

[0127] ;

[0128] Wherein, represents the average light intensity of the fish pond; represents the number of fish pond grid areas; represents the second light intensity;

[0129] In this embodiment, by combining the light intensity of each fish pond grid area and the first fish fry quantity for analysis, the second fish fry quantity is further obtained. Through effective correction, the influence of the light intensity on the fish fry quantity estimation can be avoided, and the accuracy of the fish fry quantity estimation is further improved.

[0130] In this embodiment, the fish pond is first divided into grids to obtain the first fish fry image to be detected, the light intensity, and the area of each grid area; a fish fry quantity estimation model is constructed to process the image to be detected. The gray-level co-occurrence matrix of the fish fry image to be detected is obtained through image recognition, and the texture features are obtained. Then, the first turbidity coefficient is obtained, and different denoising methods are determined according to the first turbidity coefficient; the feature extraction layer is processed by a large-scale feature extraction unit, a small-scale feature extraction unit, and an attention mechanism; the feature fusion layer obtains the first fish fry fusion feature map; the feature analysis layer uses the first fish fry fusion feature map to generate the first fish fry density map, obtains the first fish fry number, and corrects it according to the light intensity to obtain the second fish fry number; this method can effectively improve the accuracy of fish fry quantity estimation.

[0131] To verify the effectiveness of the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by the present invention, the fish fry quantity estimation accuracies of different methods are compared, including Method 1, Method 2, and Method 3; Method 1 is the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by this embodiment; Method 2 does not consider denoising on the basis of Method 1; Method 3 does not consider the influence of light intensity on the basis of Method 1. The average values of the fish fry quantity estimation accuracies of different methods for Pond B are given, and the specific results are shown in Table 2.

[0132] Table 2 Average values of fish fry quantity estimation accuracies of different methods for Pond B

[0133] Method Estimated mean accuracy Method 1 95% Method 2 85% Method 3 84%

[0134] As can be seen from Table 2, the effectiveness of the fish fry quantity estimation method based on the multi-scale semantic fusion density estimation network provided by this embodiment can effectively improve the accuracy of fish fry quantity estimation.

[0135] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for estimating the number of fry based on a multi-scale semantic fusion density estimation network, characterized in that Including: S1. Divide the fish ponds into grid areas to obtain each fish pond grid area, and acquire the first fish fry image to be detected, the second light intensity, and the third area of each fish pond grid area; S2. Construct a fish fry number estimation model to identify the first fish fry image to be detected. The fish fry number estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; the preprocessing layer performs image recognition on the first fish fry image to be detected to obtain the first turbidity coefficient; Determine different denoising methods according to the first turbidity coefficient and the preset turbidity coefficient threshold set; The process of obtaining the first turbidity coefficient includes: S10. Grayscale the first fish fry image to be detected and calculate the gray-level co-occurrence matrix of the first fish fry image to be detected; S20. Extract texture features according to the gray-level co-occurrence matrix. The texture features include contrast, uniformity, and energy; S30. Obtain the first turbidity coefficient according to the texture features; The first turbidity coefficient is: ； Among them, represents the first turbidity coefficient; represents the contrast influence weight; represents the contrast, reflecting the gray-scale difference; represents the uniformity influence weight; represents the uniformity, reflecting the texture regularity; represents the energy influence weight; represents the energy, and the energy reflects the texture smoothness; And perform multi-scale image segmentation on the first fish fry image to be detected to obtain multiple first large-scale fish fry detection images and second small-scale fish fry detection images; S3. The feature extraction layer respectively extracts features from the first large-scale fish fry detection image and the second small-scale fish fry detection image through a large-scale feature extraction unit and a small-scale feature extraction unit, and processes them through the SE attention mechanism to obtain a first large-scale feature map and a second small-scale feature map respectively; the feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain a first fish fry fusion feature map; S4. The feature analysis layer obtains a first fish fry density map according to the first fish fry fusion feature map; obtains a first fish fry number according to the first fish fry density map and the third area; and obtains a second fish fry number according to the first fish fry number, the second light intensity, and the first turbidity coefficient; the second fish fry number is: ； Among them, represents the number of second fry; represents the number of first fry; represents the weight of the influence of light intensity; represents the average light intensity of the fish pond; represents the reference value of light intensity.

2. The fry quantity estimation method based on a multi-scale semantic fusion density estimation network according to claim 1, wherein: The specific method of determining different denoising methods according to the first turbidity coefficient and the preset turbidity coefficient threshold set includes: S100. Determine the turbidity level of the fish pond grid area according to the first turbidity coefficient and the preset turbidity coefficient threshold set. The turbidity levels include mild turbidity, moderate turbidity, and severe turbidity; S200. Denoise the fish pond grid area with mild turbidity through Gaussian filtering; denoise the fish pond grid area with moderate turbidity through bilateral filtering; denoise the fish pond grid area with severe turbidity through deep learning denoising.

3. The fry quantity estimation method based on a multi-scale semantic fusion density estimation network according to claim 2, wherein: The preset turbidity coefficient threshold set includes a first turbidity coefficient threshold and a second turbidity coefficient threshold; the specific acquisition process of the turbidity level is: ； Among them, represents the turbidity level; represents mild turbidity; represents the first turbidity coefficient; represents the first turbidity coefficient threshold; represents the first turbidity coefficient threshold; represents moderate turbidity; represents severe turbidity.

4. The fish fry number estimation method based on a multi-scale semantic fusion density estimation network according to claim 1, characterized in that: The fish fry number estimation model includes an input layer, a preprocessing layer, a feature extraction layer, a feature fusion layer, a feature analysis layer, and an output layer; The preprocessing layer performs image recognition on the first fry image to be detected, obtains the first turbidity coefficient, and determines different denoising methods according to the first turbidity coefficient; and performs multi-scale image segmentation on the first fry image to be detected through U-Net to obtain a plurality of first large-scale fry detection images and second small-scale fry detection images; The feature extraction layer respectively extracts features from the first large-scale fry detection image and the second small-scale fry detection image through the large-scale feature extraction unit and the small-scale feature extraction unit, and processes them through the SE attention mechanism to obtain the first large-scale feature map and the second small-scale feature map respectively; The feature fusion layer is used to fuse the first large-scale feature map and the second small-scale feature map to obtain the first fry fusion feature map; The feature analysis layer reduces the number of channels of the first fry fusion feature map through 1*1 convolution, processes it through 3*3 convolution, and then extracts the first fry density map according to 1*1 convolution; obtains the first fry number according to the first fry density map and the third area; and obtains the second fry number according to the first fry number, the second light intensity and the first turbidity coefficient; The output layer is used to output the second fry number.

5. The fry quantity estimation method based on the multi-scale semantic fusion density estimation network according to claim 4, wherein The first fry number is: ； Wherein, represents the number of the first fry; represents the third area; represents the number of fishpond grid areas; represents the total number of pixels of the first fry density map; represents the sum of pixel values of the first fry density map.

6. The fry quantity estimation method based on a multi-scale semantic fusion density estimation network according to claim 5, characterized in that: The average light intensity of the fish pond is obtained according to the second light intensity and the number of fish pond grid areas.

Images