An intelligent screening method and system for key functional characteristic indicators of plateau fish

Abstract

The present invention discloses a method and system for intelligent screening of key functional characteristic indicators of plateau fish, and the method includes data collection, data preprocessing, plateau fish target detection, characteristic indicator prediction and estimation, and intelligent screening of key functional characteristic indicators. The present invention relates to the technical field of fish target detection, and specifically refers to a method and system for intelligent screening of key functional characteristic indicators of plateau fish. The present invention adopts a computer vision method to perform plateau fish target detection and automatically predict fish characteristic indicators, thereby improving the automation of the method and reducing damage to fish; adopts a method of visibility restoration and contrast enhancement weighted fusion to perform plateau fish image enhancement, thereby improving data availability; adopts a YOLOv8 neural network method combined with a path aggregation network to improve the overall performance of intelligent screening of key functional characteristic indicators of plateau fish.

Description

Technical Field

[0001] The present invention relates to the technical field of fish target detection, and in particular to an intelligent screening method and system for key functional characteristic indicators of plateau fish. Background Art

[0002] The intelligent screening method for key functional characteristic indicators of plateau fish aims to use computer vision technology to detect plateau fish targets and perform intelligent screening of key functional characteristic indicators of plateau fish. The traditional estimation method is represented by manual measurement, which is still widely used in most fisheries. The most commonly used method is to catch the fish for manual measurement. This measurement method belongs to contact measurement, which is usually an invasive, time-consuming and labor-intensive method. It will more or less affect the farmed fish and even cause death. The intelligent screening method for key functional characteristic indicators of plateau fish based on computer vision technology can help measure the length and height of single or multiple fish in field resource surveys and fish farming, reduce measurement time, reduce damage to fish during the collection process, and improve measurement accuracy and efficiency.

[0003] However, in the existing intelligent screening methods for key functional characteristic indicators of plateau fish, there is a technical problem that, for target detection of plateau fish, the original images detected are often of uneven quality, which reduces detection efficiency and accuracy, and thus affects the overall usability of the solution; in the existing intelligent screening methods for key functional characteristic indicators of plateau fish, there is a technical problem that plateau fish images are often small target images, and the detection performance of existing computer vision methods for small targets needs to be improved. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for intelligent screening of key functional characteristic indicators of plateau fish. In view of the fact that in the existing methods for screening key functional characteristic indicators of plateau fish, there is a traditional length estimation method represented by manual measurement, which is a contact measurement, which is usually an invasive, time-consuming and labor-intensive method, which will more or less affect the farmed fish and even cause death, and is not conducive to improving measurement efficiency and ecological protection. This solution creatively adopts computer vision methods to detect plateau fish targets and automatically predict fish characteristic indicators, thereby improving the automation of the method and reducing damage to fish; in view of the fact that in the existing methods for screening key functional characteristic indicators of plateau fish, there is a problem that target detection for plateau fish is carried out, and detection The original images often have uneven quality, which reduces detection efficiency and accuracy, and thus affects the overall usability of the solution. This solution creatively uses the visibility restoration and contrast enhancement weighted fusion method to enhance the plateau fish images, which improves the data availability and the overall accuracy of the method. In view of the technical problem that the plateau fish images are often small target images in the existing intelligent screening methods for key functional characteristic indicators of plateau fish, and the detection performance of existing computer vision methods for small targets needs to be improved, this solution creatively uses the YOLOv8 neural network method combined with the path aggregation network, and constructs a lightweight multi-scale attention subnet to perform efficient target detection of plateau fish images, thereby improving the overall performance of intelligent screening of key functional characteristic indicators of plateau fish.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent screening method for key functional characteristic indicators of plateau fish, the method comprising the following steps:

[0006] Step S1: data collection;

[0007] Step S2: data preprocessing;

[0008] Step S3: plateau fish target detection;

[0009] Step S4: feature index prediction and estimation;

[0010] Step S5: Intelligent screening of key functional characteristic indicators.

[0011] Further, in step S1, the data collection is used to collect key functional characteristic indicators of plateau fish and automatically extract the required original image data set, specifically, obtaining the original image data of plateau fish by shooting and collecting during the plateau fish survey;

[0012] The plateau fish original image data specifically refers to the side view photos of the fish.

[0013] Further, in step S2, the data preprocessing is used to perform data annotation and enhancement processing on the original image of plateau fish, specifically using an image annotation tool to perform target frame annotation, and using a fish image attribute enhancement method to perform data enhancement on the original image data of plateau fish to obtain an enhanced image data set;

[0014] The enhanced image data set specifically includes annotated text data and plateau fish enhanced image data;

[0015] The step of using an image annotation tool to annotate a target frame and performing data enhancement on the plateau fish original image data by a fish image attribute enhancement method to obtain an enhanced image data set includes:

[0016] Step S21: target frame annotation, specifically, using an image annotation tool to annotate the original image data of the plateau fish with a target frame to obtain annotated text data;

[0017] Step S22: color balance adjustment, specifically, performing color balance and white balance color correction on the plateau fish original image data to obtain a color-corrected fish image, and performing nonlinear color mapping through a kernel mapping function to obtain a nonlinearly corrected fish image;

[0018] Step S23: restoring visibility, specifically restoring the visibility of the nonlinearly corrected fish image to obtain a visibility restored fish image L1;

[0019] Step S24: contrast enhancement, specifically, contrast enhancement is performed on the nonlinearly corrected fish image to obtain a contrast enhanced fish image L2;

[0020] Step S25: image fusion, specifically, using global gradient to construct weights of visibility restored fish images and contrast enhanced fish images, obtaining visibility restored fish image fusion weights and contrast enhanced fish image fusion weights, and performing image fusion. The calculation formula for image fusion is:

[0021] Final=ω1·L1+ω2·L2;

[0022] Where Final is the fused image output, ω1 is the fusion weight of the visibility restored fish image, L1 is the visibility restored fish image, ω2 is the fusion weight of the contrast enhanced fish image, and L2 is the contrast enhanced fish image;

[0023] Step S26: enhancing the fish image attributes, specifically obtaining a fused image output through the image fusion, and obtaining plateau fish enhanced image data through obtaining the fused image output.

[0024] Further, in step S3, the plateau fish target detection is used to detect and mark plateau fish from the plateau fish image using a deep learning method, specifically, based on the enhanced image data set, a YOLOv8 neural network method combined with a path aggregation network is used to perform plateau fish target detection to obtain plateau fish target detection data, and the plateau fish target detection data specifically includes plateau fish index data and plateau fish target detection block diagram data;

[0025] The plateau fish index data specifically include the body length and height data of plateau fish;

[0026] The YOLOv8 neural network combined with the path aggregation network specifically includes a YOLOv8 basic subnet, a cross-stage biased subnet, a spatial pyramid fast subnet, a path aggregation subnet, and a lightweight multi-scale attention subnet;

[0027] The YOLOv8 basic subnet is used to build a basic model structure for target detection;

[0028] The cross-stage partial sub-network is used to reduce the number of channels and the number of convolutions;

[0029] The spatial pyramid fast subnet is used for feature map fusion;

[0030] The path aggregation subnet is used to combine the feature pyramid subnet and the pixel aggregation subnet and construct a lightweight multi-scale attention subnet;

[0031] The lightweight multi-scale attention subnet is used to introduce a multi-scale attention mechanism, improve the feature discrimination capability related to small targets in the feature map tensor, and improve the ability to obtain plateau fish target detection features;

[0032] The step of using the YOLOv8 neural network method combined with the path aggregation network to detect plateau fish targets and obtain plateau fish target detection data includes:

[0033] Step S31: constructing a YOLOv8 basic subnet, specifically constructing a depth-separable convolution block of size 1×1, generating redundant feature maps, and obtaining a feature layer of size 7×7×960 by stacking convolution blocks, performing global average pooling, and performing channel adjustment by a standard convolution block of size 1×1 to obtain a feature layer of size 1×1×1280, and using a tiling operation for full concatenation for prediction and classification, thereby constructing the YOLOv8 basic subnet;

[0034] Step S32: constructing a cross-stage partial subnet, specifically constructing the cross-stage partial subnet as a basic framework for feature extraction, and halving the number of channels through the cross-stage partial subnet to construct the cross-stage partial subnet, and obtain a cross-stage subnet feature output;

[0035] Step S33: constructing a spatial pyramid fast subnet, specifically, constructing the spatial pyramid fast subnet by adopting a downsampling technology to fuse feature maps, and obtaining a downsampling feature output;

[0036] Step S34: constructing a path aggregation subnet, specifically by constructing a feature pyramid subnet and a pixel aggregation subnet, and by aggregating the feature pyramid subnet and the pixel aggregation subnet to construct the path aggregation subnet, and obtain a path aggregation optimization feature map output;

[0037] Step S35: constructing a lightweight multi-scale attention subnet, specifically, constructing four convolutional layers through a maximum pooling operation, constructing a spatial attention sub-block, and constructing a multi-scale attention mechanism by constructing a channel attention sub-block. The multi-scale attention mechanism is used to perform feature map dimensionality reduction to obtain a multi-scale attention dimensionality reduction feature map output;

[0038] Step S36: Plateau fish target detection model training, specifically, through the YOLOv8 basic subnet, the cross-stage bias subnet, the spatial pyramid fast subnet, the path aggregation subnet and the lightweight multi-scale attention subnet, the model training is performed to obtain the plateau fish target detection model Model FD ;

[0039] Step S37: Plateau fish target detection, specifically using the plateau fish target detection model Model FD , from the enhanced image data set, plateau fish target detection data is obtained through target detection.

[0040] Further, in step S4, the characteristic index prediction and estimation is used to obtain key functional characteristic indicators of plateau fish based on the results of target detection, specifically using ImageJ software to obtain actual indicator data of plateau fish based on the plateau fish target detection frame diagram data, and by calculating the average absolute error and average relative error of the actual indicator data of plateau fish and the plateau fish indicator data, the prediction accuracy of the plateau fish characteristic indicators is judged to obtain characteristic index prediction and estimation reference data, and the characteristic index prediction and estimation reference data specifically include body length prediction accuracy reference data and body height accuracy reference data.

[0041] Furthermore, in step S5, the key functional characteristic indicators are intelligently screened for screening key functional characteristic indicators of plateau fish, specifically, based on the body length prediction accuracy reference data and the body height accuracy reference data in the characteristic indicator prediction estimation reference data, the prediction data of body length and body height of plateau fish are screened to obtain key functional characteristic indicator data of plateau fish.

[0042] The present invention provides an intelligent screening system for key functional characteristic indicators of plateau fish, comprising a data collection module, a data preprocessing module, a plateau fish target detection module, a characteristic indicator prediction and estimation module, and a key functional characteristic indicator intelligent screening module;

[0043] The data collection module is used for data collection, and obtains the original image data of plateau fish through data collection, and sends the original image data of plateau fish to the data preprocessing module;

[0044] The data preprocessing module is used for data preprocessing, and obtains an enhanced image data set through data preprocessing, and sends the enhanced image data set to the plateau fish target detection module;

[0045] The plateau fish target detection module is used for plateau fish target detection, obtains plateau fish target detection data through plateau fish target detection, and sends the plateau fish target detection data to the characteristic index prediction and estimation module;

[0046] The characteristic indicator prediction and estimation module is used for characteristic indicator prediction and estimation, obtains characteristic indicator prediction and estimation reference data through characteristic indicator prediction and estimation, and sends the characteristic indicator prediction and estimation reference data to the key function characteristic indicator intelligent screening module;

[0047] The key functional characteristic index intelligent screening module is used for intelligent screening of key functional characteristic indexes, and key functional characteristic index data of plateau fish are obtained through intelligent screening of key functional characteristic indexes.

[0048] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0049] (1) In view of the technical problem that among the existing methods for screening key functional characteristic indicators of plateau fish, the traditional length estimation method is represented by manual measurement, which is a contact measurement method. This is usually an invasive, time-consuming and labor-intensive method, which will more or less affect the farmed fish and even cause death, which is not conducive to improving measurement efficiency and ecological protection. This scheme creatively uses computer vision methods to detect plateau fish targets and automatically predict fish characteristic indicators, thereby improving the automation of the method and reducing damage to fish;

[0050] (2) In the existing intelligent screening methods for key functional characteristic indicators of plateau fish, there is a technical problem that the original images for plateau fish target detection are often of uneven quality, which reduces the detection efficiency and accuracy, and thus affects the overall usability of the solution. This solution creatively uses the visibility restoration and contrast enhancement weighted fusion method to enhance the plateau fish image, which improves the data availability and the overall accuracy of the method.

[0051] (3) In view of the technical problem that the existing intelligent screening methods for key functional characteristic indicators of plateau fish are often small target images, and the detection performance of existing computer vision methods for small targets needs to be improved, this scheme creatively adopts the YOLOv8 neural network method combined with the path aggregation network, and constructs a lightweight multi-scale attention subnet to perform efficient target detection in plateau fish images, thereby improving the overall performance of intelligent screening of key functional characteristic indicators of plateau fish. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of a process for intelligent screening of key functional characteristic indicators of plateau fish provided by the present invention;

[0053] Figure 2 A schematic diagram of an intelligent screening system for key functional characteristic indicators of plateau fish provided by the present invention;

[0054] Figure 3 This is a schematic diagram of the process of data preprocessing in step S2;

[0055] Figure 4 It is a schematic diagram of the process of plateau fish target detection in step S3;

[0056] Figure 5 is a flow chart of visibility restoration in step S23;

[0057] Figure 6 FIG. 1 is a flow chart of contrast enhancement in step S24 .

[0058] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0061] Example 1, see Figure 1 The present invention provides an intelligent screening method for key functional characteristic indicators of plateau fish, which comprises the following steps:

[0062] Step S1: data collection;

[0063] Step S2: data preprocessing;

[0064] Step S3: plateau fish target detection;

[0065] Step S4: feature index prediction and estimation;

[0066] Step S5: Intelligent screening of key functional characteristic indicators.

[0067] By performing the above operations, in view of the technical problem that in the existing methods for screening key functional characteristic indicators of plateau fish, there is a traditional length estimation method represented by manual measurement, which is a contact measurement, which is usually an invasive, time-consuming and labor-intensive method, which will more or less affect the farmed fish and even cause death, which is not conducive to improving measurement efficiency and ecological protection, this scheme creatively uses computer vision methods to detect plateau fish targets and automatically predict fish characteristic indicators, thereby improving the automation of the method and reducing damage to fish.

[0068] Example 2, see Figure 1 and Figure 2 In step S1, the data collection is used to collect key functional characteristic indicators of plateau fish and automatically extract the required original image data set, specifically, obtaining the original image data of plateau fish by shooting and collecting during the plateau fish survey;

[0069] The plateau fish original image data specifically refers to the side view photos of the fish.

[0070] Example 3, see Figure 1 , Figure 2 , Figure 3 , Figure 5 and Figure 6This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to perform data annotation and enhancement processing on the original image of plateau fish, specifically, using an image annotation tool to annotate the target frame, and using a fish image attribute enhancement method to enhance the original image data of the plateau fish to obtain an enhanced image data set;

[0071] The enhanced image data set specifically includes annotated text data and plateau fish enhanced image data;

[0072] The step of using an image annotation tool to annotate a target frame and performing data enhancement on the plateau fish original image data by a fish image attribute enhancement method to obtain an enhanced image data set includes:

[0073] Step S21: target frame annotation, specifically, using an image annotation tool to annotate the original image data of the plateau fish with a target frame to obtain annotated text data;

[0074] The image annotation tool specifically refers to Labelimg software;

[0075] The annotated text data, the specific indicator annotation format is txt format;

[0076] Step S22: color balance adjustment, specifically, performing color balance and white balance color correction on the plateau fish original image data to obtain a color-corrected fish image, and performing nonlinear color mapping through a kernel mapping function to obtain a nonlinearly corrected fish image;

[0077] Step S23: Restoring visibility, specifically restoring the visibility of the nonlinearly corrected fish image to obtain a visibility restored fish image L1, including the following steps:

[0078] Step S231: Calculate the mean value of the nonlinearly corrected fish image, and the calculation formula is:

[0079]

[0080] Where μ is the mean value of the nonlinearly corrected fish image, n is the total number of pixels of the nonlinearly corrected fish image, i is the pixel index, Z(·) is the identification function of the nonlinearly corrected fish image, and x i is the pixel of the fish image with nonlinear correction;

[0081] Step S232: Calculate the standard deviation of the nonlinearly corrected fish image, and the calculation formula is:

[0082]

[0083] where σ is the standard deviation of the nonlinearly corrected fish image, μ is the mean of the nonlinearly corrected fish image, n is the total number of pixels of the nonlinearly corrected fish image, i is the pixel index, Z(·) is the identification function of the nonlinearly corrected fish image, and x i is the pixel of the fish image with nonlinear correction;

[0084] Step S233: Calculate the upper and lower bounds of image visibility enhancement, the calculation formula is:

[0085]

[0086] Where u(x) is the upper bound of image visibility enhancement, d(x) is the lower bound of image visibility enhancement, x is the pixel identifier, Z(·) is the identification function of the nonlinearly corrected fish image, α is the visibility enhancement coefficient, ranging from [0.35, 1.25], μ is the mean of the nonlinearly corrected fish image, σ is the standard deviation of the nonlinearly corrected fish image, k is the lower bound calculation coefficient, and b is the lower bound calculation bias term;

[0087] The specific value of the visibility enhancement coefficient α is set to 1.05;

[0088] Step S234: Conjunctive normal form calculation, the calculation formula is:

[0089]

[0090] Wherein, t(x) is the calculated conjunction normal form, which is specifically calculated by the t-conorm function t(·), u(x) is the upper bound of image visibility enhancement, d(x) is the lower bound of image visibility enhancement, μ is the mean of the nonlinearly corrected fish image, and σ is the standard deviation of the nonlinearly corrected fish image;

[0091] Step S235: Calculate visibility to restore the fish image, the calculation formula is:

[0092]

[0093] Where L1 is the visibility restored fish image, max(·) is the maximum value calculation function, t(x) is the calculated conjunction normal form, and α is the visibility enhancement coefficient;

[0094] Step S24: contrast enhancement, specifically performing contrast enhancement on the nonlinearly corrected fish image to obtain a contrast enhanced fish image L2, comprising the following steps:

[0095] Step S241: Calculate the probability density of the standard normal distribution, the calculation formula is:

[0096]

[0097] Where p(·) is the probability density calculation function, x is the pixel identifier, Z(·) is the identification function of the nonlinearly corrected fish image, and exp(·) is the natural base function;

[0098] Step S242: Calculate the soft addition of the standard normal distribution, the calculation formula is:

[0099] s(x)=log(1+exp(Z(x)));

[0100] Where s(·) is the soft addition calculation function, x is the pixel identifier, Z(·) is the identification function of the nonlinear correction fish image, and exp(·) is the natural base function;

[0101] Step S243: Logarithmic image processing combined calculation, the calculation formula is:

[0102]

[0103] Where l(·) is the logarithmic image processing combination calculation function, p(·) is the probability density calculation function, x is the pixel identifier, and s(·) is the soft addition calculation function;

[0104] Step S244: Calculate the contrast-enhanced fish image, the calculation formula is:

[0105]

[0106] Where L2 is the contrast-enhanced fish image, which is calculated by a normalized function controlled by a gamma correction parameter, l(x) is the logarithmic image processing combined pixel output, which is used to represent the pixel value of the pixel point x after logarithmic image processing, min(·) is the minimum value function, max(·) is the maximum value function, and η is the gamma correction parameter used to adjust the contrast;

[0107] Step S25: image fusion, specifically, using global gradient to construct weights of visibility restored fish images and contrast enhanced fish images, obtaining visibility restored fish image fusion weights and contrast enhanced fish image fusion weights, and performing image fusion. The calculation formula for image fusion is:

[0108] Final=ω1·L1+ω2·L2;

[0109] Where Final is the fused image output, ω1 is the fusion weight of the visibility restored fish image, L1 is the visibility restored fish image, ω2 is the fusion weight of the contrast enhanced fish image, and L2 is the contrast enhanced fish image;

[0110] Step S26: enhancing the fish image attributes, specifically obtaining a fused image output through the image fusion, and obtaining plateau fish enhanced image data through obtaining the fused image output.

[0111] By performing the above operations, it is found that in the existing intelligent screening methods for key functional characteristic indicators of plateau fish, there is a technical problem that in target detection of plateau fish, the quality of the original images detected is often uneven, which reduces the detection efficiency and accuracy, and thus affects the overall usability of the solution. This solution creatively adopts the method of weighted fusion of visibility restoration and contrast enhancement to enhance the images of plateau fish, which improves the availability of data and the overall accuracy of the method.

[0112] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the plateau fish target detection is used to detect and mark plateau fish from the image of plateau fish using a deep learning method. Specifically, based on the enhanced image data set, a YOLOv8 neural network method combined with a path aggregation network is used to perform plateau fish target detection to obtain plateau fish target detection data. The plateau fish target detection data specifically includes plateau fish index data and plateau fish target detection block diagram data;

[0113] The plateau fish index data specifically include the body length and height data of plateau fish;

[0114] The YOLOv8 neural network combined with the path aggregation network specifically includes a YOLOv8 basic subnet, a cross-stage biased subnet, a spatial pyramid fast subnet, a path aggregation subnet, and a lightweight multi-scale attention subnet;

[0115] The YOLOv8 basic subnet is used to build a basic model structure for target detection;

[0116] The cross-stage partial sub-network is used to reduce the number of channels and the number of convolutions;

[0117] The spatial pyramid fast subnet is used for feature map fusion;

[0118] The path aggregation subnet is used to combine the feature pyramid subnet and the pixel aggregation subnet and construct a lightweight multi-scale attention subnet;

[0119] The lightweight multi-scale attention subnet is used to introduce a multi-scale attention mechanism, improve the feature discrimination capability related to small targets in the feature map tensor, and improve the ability to obtain plateau fish target detection features;

[0120] The step of using the YOLOv8 neural network method combined with the path aggregation network to detect plateau fish targets and obtain plateau fish target detection data includes:

[0121] Step S31: constructing a YOLOv8 basic subnet, specifically constructing a depth-separable convolution block of size 1×1, generating redundant feature maps, and obtaining a feature layer of size 7×7×960 by stacking convolution blocks, performing global average pooling, and performing channel adjustment by a standard convolution block of size 1×1 to obtain a feature layer of size 1×1×1280, and using a tiling operation for full concatenation for prediction and classification, thereby constructing the YOLOv8 basic subnet;

[0122] Step S32: constructing a cross-stage partial subnet, specifically constructing the cross-stage partial subnet as a basic framework for feature extraction, and halving the number of channels through the cross-stage partial subnet to construct the cross-stage partial subnet, and obtain a cross-stage subnet feature output;

[0123] Step S33: constructing a spatial pyramid fast subnet, specifically, constructing the spatial pyramid fast subnet by adopting a downsampling technology to fuse feature maps, and obtaining a downsampling feature output;

[0124] Step S34: constructing a path aggregation subnet, specifically by constructing a feature pyramid subnet and a pixel aggregation subnet, and by aggregating the feature pyramid subnet and the pixel aggregation subnet to construct the path aggregation subnet, and obtain a path aggregation optimization feature map output;

[0125] Step S35: constructing a lightweight multi-scale attention subnet, specifically, constructing four convolutional layers through a maximum pooling operation, constructing a spatial attention sub-block, and constructing a multi-scale attention mechanism by constructing a channel attention sub-block. The multi-scale attention mechanism is used to perform feature map dimensionality reduction to obtain a multi-scale attention dimensionality reduction feature map output;

[0126] The lightweight multi-scale subnetwork specifically uses the path aggregation optimization feature map output as a subnetwork input sample, and sends the path aggregation optimization feature map output to the spatial attention sub-block;

[0127] The spatial attention sub-block is connected to a standard convolutional layer of size 1×1, and obtains spatial attention convolutional feature data through a convolution operation, and sends the spatial attention convolutional feature data to the four-layer convolutional layer;

[0128] The four convolutional layers specifically include a first convolutional layer, a second convolutional layer, a third convolutional layer and a fourth convolutional layer;

[0129] The first convolutional layer specifically refers to a standard convolutional layer of size 1×1;

[0130] The second convolutional layer specifically includes a convolutional layer of size 1×3 and a convolutional layer of size 3×1;

[0131] The third convolutional layer specifically includes a convolutional layer of size 1×5 and a convolutional layer of size 5×1;

[0132] The fourth convolutional layer specifically includes a maximum pooling layer of size 3×3 and a convolutional layer of size 1×1;

[0133] The four convolutional layers obtain multi-scale convolution feature outputs, and obtain element-by-element multiplication feature information by performing element-by-element multiplication on the multi-scale convolution feature outputs, and send the element-by-element multiplication feature information to the channel attention sub-block;

[0134] The channel attention sub-block obtains channel attention weighted feature data by assigning weights to feature maps, and performs element-by-element addition operation on the channel attention weighted feature data and the spatial attention convolution feature data to obtain a lightweight multi-scale sub-network feature map output, and the lightweight multi-scale sub-network feature map output is used for final target detection prediction;

[0135] Step S36: Plateau fish target detection model training, specifically, through the YOLOv8 basic subnet, the cross-stage bias subnet, the spatial pyramid fast subnet, the path aggregation subnet and the lightweight multi-scale attention subnet, the model training is performed to obtain the plateau fish target detection model Model FD ;

[0136] Step S37: Plateau fish target detection, specifically using the plateau fish target detection model Model FD , from the enhanced image data set, plateau fish target detection data is obtained through target detection.

[0137] By performing the above operations, in order to address the technical problem that in the existing intelligent screening methods for key functional characteristic indicators of plateau fish, plateau fish images are often small target images, and the detection performance of existing computer vision methods for small targets needs to be improved, this scheme creatively adopts the YOLOv8 neural network method combined with the path aggregation network, and constructs a lightweight multi-scale attention subnet to perform efficient target detection of plateau fish images, thereby improving the overall performance of intelligent screening of key functional characteristic indicators of plateau fish.

[0138] Example 5, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S4, the characteristic index prediction and estimation is used to obtain the key functional characteristic index of plateau fish according to the result of target detection. Specifically, ImageJ software is used to obtain the actual index data of plateau fish according to the plateau fish target detection frame diagram data, and the prediction accuracy of the plateau fish characteristic index is judged by calculating the average absolute error and the average relative error of the actual index data of plateau fish and the index data of plateau fish, so as to obtain the characteristic index prediction and estimation reference data. The characteristic index prediction and estimation reference data specifically includes body length prediction accuracy reference data and body height accuracy reference data.

[0139] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the key functional characteristic indicators are intelligently screened to screen the key functional characteristic indicators of plateau fish. Specifically, the body length prediction accuracy reference data and the body height accuracy reference data in the characteristic indicator prediction estimation reference data are used to screen the prediction data of the body length and body height of plateau fish to obtain the key functional characteristic indicator data of plateau fish.

[0140] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides a plateau fish key functional characteristic index intelligent screening system, including a data collection module, a data preprocessing module, a plateau fish target detection module, a characteristic index prediction and estimation module and a key functional characteristic index intelligent screening module;

[0141] The data collection module is used for data collection, and obtains the original image data of plateau fish through data collection, and sends the original image data of plateau fish to the data preprocessing module;

[0142] The data preprocessing module is used for data preprocessing, and obtains an enhanced image data set through data preprocessing, and sends the enhanced image data set to the plateau fish target detection module;

[0143] The plateau fish target detection module is used for plateau fish target detection, obtains plateau fish target detection data through plateau fish target detection, and sends the plateau fish target detection data to the characteristic index prediction and estimation module;

[0144] The characteristic indicator prediction and estimation module is used for characteristic indicator prediction and estimation, obtains characteristic indicator prediction and estimation reference data through characteristic indicator prediction and estimation, and sends the characteristic indicator prediction and estimation reference data to the key function characteristic indicator intelligent screening module;

[0145] The key functional characteristic index intelligent screening module is used for intelligent screening of key functional characteristic indexes, and key functional characteristic index data of plateau fish are obtained through intelligent screening of key functional characteristic indexes.

[0146] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0147] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0148] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent screening method for key functional characteristic indicators of plateau fish, characterized by: The method comprises the following steps: Step S1: data collection, which is used to collect key functional characteristic indicators of plateau fish and automatically extract the required original image data set, specifically, obtaining the original image data of plateau fish by shooting and collecting during the plateau fish survey; Step S2: data preprocessing, which is used to perform data annotation and enhancement processing on the original images of plateau fish, specifically using an image annotation tool to annotate the target frame, and using a fish image attribute enhancement method to perform data enhancement on the original image data of plateau fish to obtain an enhanced image data set; Step S3: plateau fish target detection, which is used to detect and mark plateau fish from the plateau fish image using a deep learning method, specifically, based on the enhanced image data set, a YOLOv8 neural network method combined with a path aggregation network is used to perform plateau fish target detection to obtain plateau fish target detection data, wherein the plateau fish target detection data specifically includes plateau fish index data and plateau fish target detection block diagram data; The YOLOv8 neural network combined with the path aggregation network specifically includes a YOLOv8 basic subnet, a cross-stage biased subnet, a spatial pyramid fast subnet, a path aggregation subnet, and a lightweight multi-scale attention subnet; The YOLOv8 basic subnet is used to build a basic model structure for target detection; The cross-stage partial sub-network is used to reduce the number of channels and the number of convolutions; The spatial pyramid fast subnet is used for feature map fusion; The path aggregation subnet is used to combine the feature pyramid subnet and the pixel aggregation subnet and construct a lightweight multi-scale attention subnet; The lightweight multi-scale attention subnet is used to introduce a multi-scale attention mechanism, improve the feature discrimination capability related to small targets in the feature map tensor, and improve the ability to obtain plateau fish target detection features; The step of using the YOLOv8 neural network method combined with the path aggregation network to detect plateau fish targets and obtain plateau fish target detection data includes: Step S31: constructing a YOLOv8 basic subnet, specifically constructing a depth-separable convolution block of size 1×1, generating redundant feature maps, and obtaining a feature layer of size 7×7×960 by stacking convolution blocks, performing global average pooling, and performing channel adjustment by a standard convolution block of size 1×1 to obtain a feature layer of size 1×1×1280, and using a tiling operation for full concatenation for prediction and classification, thereby constructing the YOLOv8 basic subnet; Step S32: constructing a cross-stage partial subnet, specifically constructing the cross-stage partial subnet as a basic framework for feature extraction, and halving the number of channels through the cross-stage partial subnet to construct the cross-stage partial subnet, and obtain a cross-stage subnet feature output; Step S33: constructing a spatial pyramid fast subnet, specifically, constructing the spatial pyramid fast subnet by adopting a downsampling technology to fuse feature maps, and obtaining a downsampling feature output; Step S34: constructing a path aggregation subnet, specifically by constructing a feature pyramid subnet and a pixel aggregation subnet, and by aggregating the feature pyramid subnet and the pixel aggregation subnet to construct the path aggregation subnet, and obtain a path aggregation optimization feature map output; Step S35: constructing a lightweight multi-scale attention subnet, specifically, constructing four convolutional layers through a maximum pooling operation, constructing a spatial attention sub-block, and constructing a multi-scale attention mechanism by constructing a channel attention sub-block. The multi-scale attention mechanism is used to perform feature map dimensionality reduction to obtain a multi-scale attention dimensionality reduction feature map output; Step S36: Plateau fish target detection model training, specifically, through the YOLOv8 basic subnet, the cross-stage bias subnet, the spatial pyramid fast subnet, the path aggregation subnet and the lightweight multi-scale attention subnet, the model training is performed to obtain the plateau fish target detection model Model FD ; Step S37: Plateau fish target detection, specifically using the plateau fish target detection model Model FD , obtaining plateau fish target detection data from the enhanced image data set through target detection; Step S4: characteristic index prediction and estimation, which is used to obtain key functional characteristic indicators of plateau fish according to the results of target detection, and obtain characteristic index prediction and estimation reference data; Step S5: Intelligent screening of key functional characteristic indicators, which is used to screen the key functional characteristic indicators of plateau fish to obtain key functional characteristic indicator data of plateau fish.

2. The intelligent screening method for key functional characteristic indicators of plateau fish according to claim 1 is characterized by: In step S2, the enhanced image data set specifically includes annotated text data and plateau fish enhanced image data; The step of using an image annotation tool to annotate a target frame and performing data enhancement on the plateau fish original image data by a fish image attribute enhancement method to obtain an enhanced image data set includes: Step S21: target frame annotation, specifically, using an image annotation tool to annotate the original image data of the plateau fish with a target frame to obtain annotated text data; Step S22: color balance adjustment, specifically, performing color balance and white balance color correction on the plateau fish original image data to obtain a color-corrected fish image, and performing nonlinear color mapping through a kernel mapping function to obtain a nonlinearly corrected fish image; Step S23: restoring visibility, specifically restoring the visibility of the nonlinearly corrected fish image to obtain a visibility restored fish image L1; Step S24: contrast enhancement, specifically performing contrast enhancement on the nonlinearly corrected fish image to obtain a contrast enhanced fish image L2; Step S25: image fusion, specifically, using global gradient to construct weights of visibility restored fish images and contrast enhanced fish images, obtaining visibility restored fish image fusion weights and contrast enhanced fish image fusion weights, and performing image fusion. The calculation formula for image fusion is: Final=ω1·L1+ω2·L2; Where Final is the fused image output, ω1 is the fusion weight of the visibility restored fish image, L1 is the visibility restored fish image, ω2 is the fusion weight of the contrast enhanced fish image, and L2 is the contrast enhanced fish image; Step S26: enhancing the fish image attributes, specifically obtaining a fused image output through the image fusion, and obtaining plateau fish enhanced image data through obtaining the fused image output.

3. The intelligent screening method for key functional characteristic indicators of plateau fish according to claim 2 is characterized by: In step S4, the characteristic index prediction and estimation is used to obtain the key functional characteristic index of plateau fish according to the result of target detection. Specifically, ImageJ software is used to obtain the actual index data of plateau fish according to the plateau fish target detection frame diagram data, and the prediction accuracy of the plateau fish characteristic index is judged by calculating the average absolute error and the average relative error of the actual index data of plateau fish and the index data of plateau fish, so as to obtain the characteristic index prediction and estimation reference data. The characteristic index prediction and estimation reference data specifically include body length prediction accuracy reference data and body height accuracy reference data.

4. The intelligent screening method for key functional characteristic indicators of plateau fish according to claim 3 is characterized by: In step S5, the key functional characteristic indicators are intelligently screened to screen the key functional characteristic indicators of plateau fish. Specifically, the predicted data of body length and height of plateau fish are screened based on the body length prediction accuracy reference data and body height accuracy reference data in the characteristic indicator prediction estimation reference data to obtain the key functional characteristic indicator data of plateau fish.

5. The intelligent screening method for key functional characteristic indicators of plateau fish according to claim 4 is characterized by: In step S1, the plateau fish original image data specifically refers to a side view photo of the fish.

6. An intelligent screening system for key functional characteristic indicators of plateau fish, used to implement an intelligent screening method for key functional characteristic indicators of plateau fish as described in any one of claims 1 to 5, characterized in that: It includes a data collection module, a data preprocessing module, a plateau fish target detection module, a characteristic index prediction and estimation module and a key functional characteristic index intelligent screening module.

7. The intelligent screening system for key functional characteristic indicators of plateau fish according to claim 6 is characterized by: The data collection module is used for data collection, and obtains the original image data of plateau fish through data collection, and sends the original image data of plateau fish to the data preprocessing module; The data preprocessing module is used for data preprocessing, and obtains an enhanced image data set through data preprocessing, and sends the enhanced image data set to the plateau fish target detection module; The plateau fish target detection module is used for plateau fish target detection, obtains plateau fish target detection data through plateau fish target detection, and sends the plateau fish target detection data to the characteristic index prediction and estimation module; The characteristic indicator prediction and estimation module is used for characteristic indicator prediction and estimation, obtains characteristic indicator prediction and estimation reference data through characteristic indicator prediction and estimation, and sends the characteristic indicator prediction and estimation reference data to the key function characteristic indicator intelligent screening module; The key functional characteristic index intelligent screening module is used for intelligent screening of key functional characteristic indexes, and key functional characteristic index data of plateau fish are obtained through intelligent screening of key functional characteristic indexes.

Images