Precision phenotypic analysis method and system of Leopard gill perch based on deep learning

Abstract

The present invention discloses a method and system for accurately analyzing the phenotype of leopard gill perch based on deep learning. The key frame images of leopard gill perch are obtained through video monitoring, and the deep learning CNN model is used to identify fish schools, segment fish image areas and extract phenotypic features. A breeding cycle is selected, and the image area of ​​leopard gill perch is divided into sub-areas. The phenotypic features and preset variation feature data are calculated for feature differences to obtain variation characteristic values, and weights are assigned to each sub-area. The system stores the image and feature data of each identified target. In real-time fish school variation identification and statistical analysis, similarity calculations are performed between real-time targets and existing targets in the database to achieve repeatability detection and accurate target tracking of leopard gill perch. Through the present invention, rapid and accurate phenotypic analysis and target tracking of leopard gill perch can be achieved, and the accuracy of breeding analysis and breeding efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the field of fish phenotypic analysis, and more specifically, to a method and system for accurately analyzing the phenotype of leopard spiny bass based on deep learning. Background Art

[0002] As an important economic fish, the healthy development of the aquaculture industry of Leopard gill perch is of great significance to improving the quality of aquatic resources. However, traditional aquaculture management methods and phenotyping processes often rely on manual observation and empirical judgment, resulting in low aquaculture efficiency and difficulty in accurately assessing the growth status of fish populations and the diversity of genetic phenotypes. Therefore, how to achieve accurate and rapid analysis of the phenotype of Leopard gill perch is an important issue that needs to be solved urgently. Summary of the invention

[0003] The present invention overcomes the defects of the prior art and proposes a method and system for accurately analyzing the phenotype of leopard gill perch based on deep learning.

[0004] The first aspect of the present invention provides a method for accurately analyzing the phenotype of leopard gill perch based on deep learning, comprising:

[0005] During multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set;

[0006] A CNN image analysis model based on deep learning was constructed, and the key image set was imported into the image analysis model for fish school recognition, fish region segmentation, and fish phenotypic feature extraction. Feature extraction included contour texture features, color features, and texture distribution features, and the target image region and target feature data based on the leopard spiny perch image were obtained.

[0007] A breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the feature difference calculation based on the standard Euclidean distance is performed on the current target feature data and the preset variation feature data. The difference calculation process is calculated based on each sub-area separately, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted, and the corresponding weight is assigned to each sub-area based on the variation characteristic value;

[0008] In the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database;

[0009] In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain the similarity index. The similarity index is used to perform repeatability detection on the real-time identification target and track the identification target;

[0010] In each breeding cycle, the fish school information and mutation identification results are sent to the preset terminal device.

[0011] In this scheme, during multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set, specifically:

[0012] Set the breeding cycle, set up multiple monitoring points in the target breeding area, and set up video monitoring devices at each monitoring point;

[0013] Use video monitoring devices to monitor the target breeding area and obtain video data of leopard gill perch;

[0014] Key frames are extracted from video data to obtain key image sets, which are then preprocessed by image denoising, enhancement, and standardization.

[0015] In this scheme, the CNN image analysis model based on deep learning is constructed, and the key image set is imported into the image analysis model for fish school identification, fish area segmentation and fish phenotypic feature extraction. The feature extraction includes the extraction of contour texture features, color features, and texture distribution features, and the target image area and target feature data based on the leopard gill perch image are obtained, specifically:

[0016] The CNN image analysis model based on deep learning includes convolutional layer, pooling layer and fully connected layer;

[0017] Based on historical image data, the image analysis model was trained to extract features and identify leopard gill perch;

[0018] Import the key image set into the image analysis model for fish school recognition, perform fish school recognition through each image, and perform fish area segmentation on the leopard gill perch based on the recognition results to obtain the target image area;

[0019] Image segmentation is performed based on the target image area and based on the image analysis model, the convolution layer and the pooling layer are used to perform convolution operations on the input image and extract local edge and texture features to obtain contour texture features;

[0020] Performing color statistics on the target image area by means of a color histogram and generating corresponding color feature data based on the color statistical data in the histogram;

[0021] The body color and pattern distribution is analyzed through contour texture features and color feature data to form texture distribution features based on body color and pattern.

[0022] In this solution, a breeding cycle is selected, the current identified fish information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the current target feature data and the preset variation feature data are subjected to feature difference calculation based on the standard Euclidean distance. The difference calculation process is based on each sub-area, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area. The variation characteristic values ​​are sorted and the corresponding weights are assigned to each sub-area based on the variation characteristic values, specifically:

[0023] A breeding cycle is selected, and a key image set is acquired and the identification and analysis of the leopard gill perch are performed on the breeding cycle to obtain current identification fish school information;

[0024] By using the current identified fish school information, the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively;

[0025] For the current target image area, each body part of the leopard gill perch is identified and segmented to obtain multiple sub-areas;

[0026] Taking each sub-region as the analysis unit, extract the feature data of the sub-region from the current target feature data and convert it into a feature vector with the preset variation feature data, and calculate the feature difference based on the standard Euclidean distance to obtain the feature difference value of each sub-region;

[0027] The variation characteristic value of each sub-region is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted and the corresponding weight is assigned to each sub-region based on the variation characteristic value.

[0028] In this solution, in the selected breeding cycle, based on the current identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database, specifically:

[0029] According to the current fish school identification information, the target image area corresponding to each identification target is mapped and associated with the target feature data to form a data table and store it in the system database.

[0030] In this scheme, in each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. In the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain a similarity index. The real-time identification target is detected for repeatability and the identification target is tracked and marked through the similarity index, specifically:

[0031] In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked;

[0032] Mark the target feature data of the real-time recognition target as the first feature data, and select a target feature data corresponding to the recognition target from the system database and mark it as the second feature data;

[0033] The first feature data and the second feature data are similarly calculated. The calculation process is based on each sub-region. The similarity calculation is performed by converting the first feature data and the second feature data into feature vectors and calculating the similarity based on the standard Euclidean distance to obtain M similarities, where M is the number of sub-regions. Based on the corresponding weights, the M similarities are weighted averaged to obtain a similarity index;

[0034] Determine whether the similarity index is greater than a preset similarity, and if so, determine that the real-time recognition target and the selected recognition target are the same target;

[0035] Based on the similarity index, the real-time recognition target is detected for repeatability and the recognition target is tracked and marked.

[0036] In this solution, in each breeding cycle, the fish school information and the variation identification results are sent to a preset terminal device, which includes a mobile terminal and a computer terminal.

[0037] The second aspect of the present invention further provides a system for accurately analyzing the phenotype of leopard gill perch based on deep learning, the system comprising: a memory and a processor, the memory comprising a program for accurately analyzing the phenotype of leopard gill perch based on deep learning, and the program for accurately analyzing the phenotype of leopard gill perch based on deep learning is executed by the processor to implement the following steps:

[0038] During multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set;

[0039] A CNN image analysis model based on deep learning was constructed, and the key image set was imported into the image analysis model for fish school recognition, fish region segmentation, and fish phenotypic feature extraction. Feature extraction included contour texture features, color features, and texture distribution features, and the target image region and target feature data based on the leopard spiny perch image were obtained.

[0040] A breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the feature difference calculation based on the standard Euclidean distance is performed on the current target feature data and the preset variation feature data. The difference calculation process is calculated based on each sub-area separately, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted, and the corresponding weight is assigned to each sub-area based on the variation characteristic value;

[0041] In the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database;

[0042] In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain the similarity index. The similarity index is used to perform repeatability detection on the real-time identification target and track the identification target;

[0043] In each breeding cycle, the fish school information and mutation identification results are sent to the preset terminal device.

[0044] The third aspect of the present invention also provides a computer-readable storage medium, which includes a program for precise phenotypic analysis of leopard gill perch based on deep learning. When the program for precise phenotypic analysis of leopard gill perch based on deep learning is executed by a processor, the steps of the method for precise phenotypic analysis of leopard gill perch based on deep learning as described in any of the above items are implemented.

[0045] The present invention discloses a method and system for accurately analyzing the phenotype of leopard gill perch based on deep learning. The key frame images of leopard gill perch are obtained through video monitoring, and the deep learning CNN model is used to identify fish schools, segment fish image areas and extract phenotypic features. A breeding cycle is selected, and the image area of ​​leopard gill perch is divided into sub-areas. The phenotypic features and preset variation feature data are calculated for feature differences to obtain variation characteristic values, and weights are assigned to each sub-area. The system stores the image and feature data of each identified target. In real-time fish school variation identification and statistical analysis, similarity calculations are performed between real-time targets and existing targets in the database to achieve repeatability detection and accurate target tracking of leopard gill perch. Through the present invention, rapid and accurate phenotypic analysis and target tracking of leopard gill perch can be achieved, and the accuracy of breeding analysis and breeding efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flow chart of a method for accurately analyzing the phenotype of leopard gill perch based on deep learning of the present invention is shown;

[0047] Figure 2 A block diagram of a deep learning-based precise phenotype analysis system for leopard gill perch of the present invention is shown. DETAILED DESCRIPTION

[0048] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0050] Figure 1 A flow chart of a method for accurately analyzing the phenotype of leopard gill perch based on deep learning according to the present invention is shown.

[0051] like Figure 1 As shown, the first aspect of the present invention provides a method for accurately analyzing the phenotype of leopard gill perch based on deep learning, comprising:

[0052] S102, monitoring the target breeding area through a video monitoring device during multiple breeding cycles, obtaining video data of the leopard gill perch, extracting key frames from the video data, and obtaining a key image set;

[0053] S104, constructing a CNN image analysis model based on deep learning, importing the key image set into the image analysis model to perform fish school recognition, fish region segmentation and fish phenotypic feature extraction, wherein the feature extraction includes extraction of contour texture features, color features and texture distribution features, and obtaining target image regions and target feature data based on the leopard spiny perch image;

[0054] S106, selecting a breeding cycle, obtaining the current identified fish school information, and marking the corresponding target image area and target feature data as the current target image area and the current target feature data, respectively, dividing the current target image area into multiple sub-areas, performing feature difference calculation based on the standard Euclidean distance between the current target feature data and the preset variation feature data, the difference calculation process is calculated based on each sub-area, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted, and a corresponding weight is assigned to each sub-area based on the variation characteristic value;

[0055] S108, in the selected breeding cycle, based on the currently identified fish school information, storing the target image area and target feature data corresponding to each identified target in the system database;

[0056] S110, in each breeding cycle, fish school variation recognition and statistics are performed through the target image area and the target feature data. In the statistical process, a real-time recognition target is marked, and similarity calculation is performed based on the target feature data of the real-time recognition target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area to perform comprehensive calculation to obtain a similarity index. The real-time recognition target is detected for repeatability and the recognition target is tracked and marked through the similarity index;

[0057] S112, in each breeding cycle, the fish school information and the variation identification results are sent to a preset terminal device.

[0058] It should be noted that the video monitoring device is a high-definition video monitoring device, which is used to collect high-definition images of the target breeding area to analyze the phenotype of the fish school.

[0059] According to an embodiment of the present invention, during multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set, specifically:

[0060] Set the breeding cycle, set up multiple monitoring points in the target breeding area, and set up video monitoring devices at each monitoring point;

[0061] Use video monitoring devices to monitor the target breeding area and obtain video data of leopard gill perch;

[0062] Key frames are extracted from video data to obtain key image sets, which are then preprocessed by image denoising, enhancement, and standardization.

[0063] It should be noted that the breeding cycle is set by the user and is used to study the growth variation of leopard gill perch in different time periods, which can provide data support for genetic diversity, breeding analysis, breeding regulation and other research. For the key image set, the corresponding atlas analysis is performed based on each breeding cycle, specifically the target recognition, image segmentation, feature extraction and other analysis processes of leopard gill perch.

[0064] According to an embodiment of the present invention, the CNN image analysis model based on deep learning is constructed, and the key image set is imported into the image analysis model for fish school identification, fish region segmentation and fish phenotypic feature extraction, and the feature extraction includes the extraction of contour texture features, color features, and texture distribution features, and the target image area and target feature data based on the leopard gill perch image are obtained, specifically:

[0065] The CNN image analysis model based on deep learning includes convolutional layer, pooling layer and fully connected layer;

[0066] Based on historical image data, the image analysis model was trained to extract features and identify leopard gill perch;

[0067] Import the key image set into the image analysis model for fish school recognition, perform fish school recognition through each image, and perform fish area segmentation on the leopard gill perch based on the recognition results to obtain the target image area;

[0068] Image segmentation is performed based on the target image area and based on the image analysis model, the convolution layer and the pooling layer are used to perform convolution operations on the input image and extract local edge and texture features to obtain contour texture features;

[0069] Performing color statistics on the target image area by means of a color histogram and generating corresponding color feature data based on the color statistical data in the histogram;

[0070] The body color and pattern distribution is analyzed through contour texture features and color feature data to form texture distribution features based on body color and pattern.

[0071] It should be noted that the image analysis model mainly consists of a convolution layer, a pooling layer, a fully connected layer, etc., which are used for feature extraction. The historical image data is the image data of the leopard gill perch. Contour texture is the contour and texture features. In the target image area and the target feature data, it is generally obtained based on a leopard gill perch. Therefore, in the image set analysis, the analysis data of multiple leopard gill perches are included, that is, the analysis result includes multiple target image areas and corresponding multiple target feature data. In the process of variation analysis, variation analysis is generally performed on a single fish, and the variation of the entire fish school is gradually analyzed. Color feature data can be used for body color analysis and color distribution analysis. Texture features can be used for pattern analysis of leopard gill perch.

[0072] According to an embodiment of the present invention, a breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the current target feature data and the preset variation feature data are subjected to feature difference calculation based on the standard Euclidean distance. The difference calculation process is based on each sub-area, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area. The variation characteristic values ​​are sorted and the corresponding weights are assigned to each sub-area based on the variation characteristic values, specifically:

[0073] A breeding cycle is selected, and a key image set is acquired and the identification and analysis of the leopard gill perch are performed on the breeding cycle to obtain current identification fish school information;

[0074] By using the current identified fish school information, the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively;

[0075] For the current target image area, each body part of the leopard gill perch is identified and segmented to obtain multiple sub-areas;

[0076] Taking each sub-region as the analysis unit, extract the feature data of the sub-region from the current target feature data and convert it into a feature vector with the preset variation feature data, and calculate the feature difference based on the standard Euclidean distance to obtain the feature difference value of each sub-region;

[0077] The variation characteristic value of each sub-region is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted and the corresponding weight is assigned to each sub-region based on the variation characteristic value.

[0078] It should be noted that the target identification in the present invention specifically refers to the leopard gill perch object. The fish school identification information includes fish school identification information, the number of identification targets, the target image area corresponding to the identification target and the target feature data, etc. The data are stored in the system database, and each breeding cycle corresponds to the fish school identification information. In the present invention, the sub-region division helps to divide different morphological traits into regions and analyze them independently, such as the variation analysis of phenotypes such as head length, eye diameter, tail peduncle length, and fish body color.

[0079] The characteristic difference value calculation formula is as follows:

[0080] T = K1 × D;

[0081] Wherein, K1 represents the correction coefficient, D is the distance value between the corresponding feature data and the preset variation feature data in a sub-region, the distance value is calculated using the standard Euclidean distance method, and T is the feature difference value.

[0082] The feature difference value represents the degree of variation. The larger the value, the greater the degree of variation.

[0083] The formula for calculating the variation characteristic value is as follows:

[0084] ,

[0085] in, represents the variation characteristic value of a sub-region, N is the number of recognition targets, is the correction factor, is the feature difference value corresponding to a sub-region in the i-th recognition target.

[0086] It is worth mentioning that the number of identified targets is the value obtained by target identification and statistics through image sets within a selected breeding cycle. The selected breeding cycle is generally the first cycle, which is used for preliminary identification and variation analysis of farmed fish in the current target breeding area. Most of them are identified through multiple monitoring points. Since it is a preliminary identification, there is a possibility of repeated target detection. The purpose of preliminary identification is to make preliminary statistics on the variability of the entire fish school, that is, most of the fish schools are counted, and weight analysis and assignment are performed on different sub-areas of the identified targets, further providing a data basis for subsequent target tracking and repeatability detection. The sub-areas are sorted by the variation characteristic value. The larger the variation characteristic value, the greater the weight of the corresponding sub-area.

[0087] According to an embodiment of the present invention, in the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database, specifically:

[0088] According to the current fish school identification information, the target image area corresponding to each identification target is mapped and associated with the target feature data to form a data table and store it in the system database.

[0089] It should be noted that the role of mapping association is to establish an association relationship for the corresponding data of each identification target, which can improve data processing efficiency during subsequent data search and screening.

[0090] According to an embodiment of the present invention, in each breeding cycle, fish school variation identification and statistics are performed through the target image area and the target feature data. In the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain a similarity index. The real-time identification target is detected for repeatability and the identification target is tracked and marked through the similarity index, specifically:

[0091] In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked;

[0092] Mark the target feature data of the real-time recognition target as the first feature data, and select a target feature data corresponding to the recognition target from the system database and mark it as the second feature data;

[0093] The first feature data and the second feature data are similarly calculated. The calculation process is based on each sub-region. The similarity calculation is performed by converting the first feature data and the second feature data into feature vectors and calculating the similarity based on the standard Euclidean distance to obtain M similarities, where M is the number of sub-regions. Based on the corresponding weights, the M similarities are weighted averaged to obtain a similarity index;

[0094] Determine whether the similarity index is greater than a preset similarity, and if so, determine that the real-time recognition target and the selected recognition target are the same target;

[0095] Based on the similarity index, the real-time recognition target is detected for repeatability and the recognition target is tracked and marked.

[0096] It should be noted that the similarity calculation is performed by converting the first feature data and the second feature data into feature vectors, and performing similarity calculation based on the standard Euclidean distance. The feature vector conversion and similarity calculation can be performed based on contour texture features, color features, etc. The similarity is determined according to the distance value, and the larger the distance value, the smaller the similarity.

[0097] The first feature data and the second feature data are overall features of an object, and the similarity analysis process is to perform similarity calculation based on the features of each sub-region.

[0098] In the large-scale breeding of leopard gill perch and phenotypic analysis of leopard gill perch, the prior art is mostly based on artificial phenotypic observation and recording, and the process is time-consuming and laborious. Based on this, the present invention collects the characteristic data of most of the leopard gill perch in the breeding area by setting an initial cycle, and performs variation characteristic analysis, assigns the corresponding weight of each sub-region based on the variation characteristic value, and performs similarity index calculation in the subsequent process by the weight, for repeatability judgment and precision tracking of the target. The larger the variation characteristic value of a sub-region, the larger the discrimination degree, and therefore, the larger the corresponding weight. Through the present invention, it is possible to effectively and quickly detect whether the real-time target exists in the existing identification target, and perform analysis and recording of phenotypic characteristics. If it is judged that there is a repeated identification target in a breeding cycle, skip detection can be performed, thereby ensuring that the phenotypic characteristic analysis, recording and real-time tracking of the target of non-repetition in the image set are performed, improving the efficiency and accuracy of phenotypic analysis, realizing rapid analysis and precision recording of the target phenotype in the large-scale breeding process, reducing manual participation, realizing informatization and automatic breeding analysis, and improving breeding efficiency.

[0099] According to an embodiment of the present invention, in each breeding cycle, the fish school information and the variation identification result are sent to a preset terminal device, and the preset terminal device includes a mobile terminal and a computer terminal.

[0100] It should be noted that the variation identification results include the comparison results of the target feature data of the real-time target with the target feature data of the previous cycle and the comparison results of the target feature data of the real-time target with the preset variation feature data. Through real-time target detection and tracking analysis, the overall phenotypic variation analysis of the leopard gill perch in the breeding area can be carried out, and the variation feature comparison can be used to analyze the impact of phenotype and body mass based on multivariate regression analysis.

[0101] Figure 2 A block diagram of a deep learning-based precise phenotype analysis system for leopard gill perch of the present invention is shown.

[0102] The second aspect of the present invention further provides a system 2 for accurately analyzing the phenotype of leopard gill perch based on deep learning, the system comprising: a memory 21 and a processor 22, wherein the memory 21 comprises a program for accurately analyzing the phenotype of leopard gill perch based on deep learning, and the program for accurately analyzing the phenotype of leopard gill perch based on deep learning is executed by the processor 22 to implement the following steps:

[0103] During multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set;

[0104] A CNN image analysis model based on deep learning was constructed, and the key image set was imported into the image analysis model for fish school recognition, fish region segmentation, and fish phenotypic feature extraction. Feature extraction included contour texture features, color features, and texture distribution features, and the target image region and target feature data based on the leopard spiny perch image were obtained.

[0105] A breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the feature difference calculation based on the standard Euclidean distance is performed on the current target feature data and the preset variation feature data. The difference calculation process is calculated based on each sub-area separately, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted, and the corresponding weight is assigned to each sub-area based on the variation characteristic value;

[0106] In the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database;

[0107] In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain the similarity index. The similarity index is used to perform repeatability detection on the real-time identification target and track the identification target;

[0108] In each breeding cycle, the fish school information and mutation identification results are sent to the preset terminal device.

[0109] It should be noted that the video monitoring device is a high-definition video monitoring device, which is used to collect high-definition images of the target breeding area to analyze the phenotype of the fish school.

[0110] According to an embodiment of the present invention, during multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set, specifically:

[0111] Set the breeding cycle, set up multiple monitoring points in the target breeding area, and set up video monitoring devices at each monitoring point;

[0112] Use video monitoring devices to monitor the target breeding area and obtain video data of leopard gill perch;

[0113] Key frames are extracted from video data to obtain key image sets, which are then preprocessed by image denoising, enhancement, and standardization.

[0114] It should be noted that the breeding cycle is set by the user and is used to study the growth variation of leopard gill perch in different time periods, which can provide data support for genetic diversity, breeding analysis, breeding regulation and other research. For the key image set, the corresponding atlas analysis is performed based on each breeding cycle, specifically the target recognition, image segmentation, feature extraction and other analysis processes of leopard gill perch.

[0115] According to an embodiment of the present invention, the CNN image analysis model based on deep learning is constructed, and the key image set is imported into the image analysis model for fish school identification, fish region segmentation and fish phenotypic feature extraction, and the feature extraction includes the extraction of contour texture features, color features, and texture distribution features, and the target image area and target feature data based on the leopard gill perch image are obtained, specifically:

[0116] The CNN image analysis model based on deep learning includes convolutional layer, pooling layer and fully connected layer;

[0117] Based on historical image data, the image analysis model was trained to extract features and identify leopard gill perch;

[0118] Import the key image set into the image analysis model for fish school recognition, perform fish school recognition through each image, and perform fish area segmentation on the leopard gill perch based on the recognition results to obtain the target image area;

[0119] Image segmentation is performed based on the target image area and based on the image analysis model, the convolution layer and the pooling layer are used to perform convolution operations on the input image and extract local edge and texture features to obtain contour texture features;

[0120] Performing color statistics on the target image area by means of a color histogram and generating corresponding color feature data based on the color statistical data in the histogram;

[0121] The body color and pattern distribution is analyzed through contour texture features and color feature data to form texture distribution features based on body color and pattern.

[0122] It should be noted that the image analysis model mainly consists of a convolution layer, a pooling layer, a fully connected layer, etc., which are used for feature extraction. The historical image data is the image data of the leopard gill perch. Contour texture is the contour and texture features. In the target image area and the target feature data, it is generally obtained based on a leopard gill perch. Therefore, in the image set analysis, the analysis data of multiple leopard gill perches are included, that is, the analysis result includes multiple target image areas and corresponding multiple target feature data. In the process of variation analysis, variation analysis is generally performed on a single fish, and the variation of the entire fish school is gradually analyzed. Color feature data can be used for body color analysis and color distribution analysis. Texture features can be used for pattern analysis of leopard gill perch.

[0123] According to an embodiment of the present invention, a breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the current target feature data and the preset variation feature data are subjected to feature difference calculation based on the standard Euclidean distance. The difference calculation process is based on each sub-area, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area. The variation characteristic values ​​are sorted and the corresponding weights are assigned to each sub-area based on the variation characteristic values, specifically:

[0124] A breeding cycle is selected, and a key image set is acquired and the identification and analysis of the leopard gill perch are performed on the breeding cycle to obtain current identification fish school information;

[0125] By using the current identified fish school information, the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively;

[0126] For the current target image area, each body part of the leopard gill perch is identified and segmented to obtain multiple sub-areas;

[0127] Taking each sub-region as the analysis unit, extract the feature data of the sub-region from the current target feature data and convert it into a feature vector with the preset variation feature data, and calculate the feature difference based on the standard Euclidean distance to obtain the feature difference value of each sub-region;

[0128] The variation characteristic value of each sub-region is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted and the corresponding weight is assigned to each sub-region based on the variation characteristic value.

[0129] It should be noted that the target identification in the present invention specifically refers to the leopard gill perch object. The fish school identification information includes fish school identification information, the number of identification targets, the target image area corresponding to the identification target and the target feature data, etc. The data are stored in the system database, and each breeding cycle corresponds to the fish school identification information. In the present invention, the sub-region division helps to divide different morphological traits into regions and analyze them independently, such as the variation analysis of phenotypes such as head length, eye diameter, tail peduncle length, and fish body color.

[0130] The characteristic difference value calculation formula is as follows:

[0131] T = K1 × D;

[0132] Wherein, K1 represents the correction coefficient, D is the distance value between the corresponding feature data and the preset variation feature data in a sub-region, the distance value is calculated using the standard Euclidean distance method, and T is the feature difference value.

[0133] The feature difference value represents the degree of variation. The larger the value, the greater the degree of variation.

[0134] The formula for calculating the variation characteristic value is as follows:

[0135] ,

[0136] in, represents the variation characteristic value of a sub-region, N is the number of recognition targets, is the correction factor, is the feature difference value corresponding to a sub-region in the i-th recognition target.

[0137] It is worth mentioning that the number of identified targets is the value obtained by target identification and statistics through image sets within a selected breeding cycle. The selected breeding cycle is generally the first cycle, which is used for preliminary identification and variation analysis of farmed fish in the current target breeding area. Most of them are identified through multiple monitoring points. Since it is a preliminary identification, there is a possibility of repeated target detection. The purpose of preliminary identification is to make preliminary statistics on the variability of the entire fish school, that is, most of the fish schools are counted, and weight analysis and assignment are performed on different sub-areas of the identified targets, further providing a data basis for subsequent target tracking and repeatability detection. The sub-areas are sorted by the variation characteristic value. The larger the variation characteristic value, the greater the weight of the corresponding sub-area.

[0138] According to an embodiment of the present invention, in the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database, specifically:

[0139] According to the current fish school identification information, the target image area corresponding to each identification target is mapped and associated with the target feature data to form a data table and store it in the system database.

[0140] It should be noted that the role of mapping association is to establish an association relationship for the corresponding data of each identification target, which can improve data processing efficiency during subsequent data search and screening.

[0141] According to an embodiment of the present invention, in each breeding cycle, fish school variation identification and statistics are performed through the target image area and the target feature data. In the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain a similarity index. The real-time identification target is detected for repeatability and the identification target is tracked and marked through the similarity index, specifically:

[0142] In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked;

[0143] Mark the target feature data of the real-time recognition target as the first feature data, and select a target feature data corresponding to the recognition target from the system database and mark it as the second feature data;

[0144] The first feature data and the second feature data are similarly calculated. The calculation process is based on each sub-region. The similarity calculation is performed by converting the first feature data and the second feature data into feature vectors and calculating the similarity based on the standard Euclidean distance to obtain M similarities, where M is the number of sub-regions. Based on the corresponding weights, the M similarities are weighted averaged to obtain a similarity index;

[0145] Determine whether the similarity index is greater than a preset similarity, and if so, determine that the real-time recognition target and the selected recognition target are the same target;

[0146] Based on the similarity index, the real-time recognition target is detected for repeatability and the recognition target is tracked and marked.

[0147] It should be noted that the similarity calculation is performed by converting the first feature data and the second feature data into feature vectors, and performing similarity calculation based on the standard Euclidean distance. The feature vector conversion and similarity calculation can be performed based on contour texture features, color features, etc. The similarity is determined according to the distance value, and the larger the distance value, the smaller the similarity.

[0148] The first feature data and the second feature data are overall features of an object, and the similarity analysis process is to perform similarity calculation based on the features of each sub-region.

[0149] In the large-scale breeding of leopard gill perch and phenotypic analysis of leopard gill perch, the prior art is mostly based on artificial phenotypic observation and recording, and the process is time-consuming and laborious. Based on this, the present invention collects the characteristic data of most of the leopard gill perch in the breeding area by setting an initial cycle, and performs variation characteristic analysis, assigns the corresponding weight of each sub-region based on the variation characteristic value, and performs similarity index calculation in the subsequent process by the weight, for repeatability judgment and precision tracking of the target. The larger the variation characteristic value of a sub-region, the larger the discrimination degree, and therefore, the larger the corresponding weight. Through the present invention, it is possible to effectively and quickly detect whether the real-time target exists in the existing identification target, and perform analysis and recording of phenotypic characteristics. If it is judged that there is a repeated identification target in a breeding cycle, skip detection can be performed, thereby ensuring that the phenotypic characteristic analysis, recording and real-time tracking of the target of non-repetition in the image set are performed, improving the efficiency and accuracy of phenotypic analysis, realizing rapid analysis and precision recording of the target phenotype in the large-scale breeding process, reducing manual participation, realizing informatization and automatic breeding analysis, and improving breeding efficiency.

[0150] According to an embodiment of the present invention, in each breeding cycle, the fish school information and the variation identification result are sent to a preset terminal device, and the preset terminal device includes a mobile terminal and a computer terminal.

[0151] It should be noted that the variation identification results include the comparison results of the target feature data of the real-time target with the target feature data of the previous cycle and the comparison results of the target feature data of the real-time target with the preset variation feature data. Through real-time target detection and tracking analysis, the overall phenotypic variation analysis of the leopard gill perch in the breeding area can be carried out, and the variation feature comparison can be used to analyze the impact of phenotype and body mass based on multivariate regression analysis.

[0152] The third aspect of the present invention also provides a computer-readable storage medium, which includes a program for precise phenotypic analysis of leopard gill perch based on deep learning. When the program for precise phenotypic analysis of leopard gill perch based on deep learning is executed by a processor, the steps of the method for precise phenotypic analysis of leopard gill perch based on deep learning as described in any of the above items are implemented.

[0153] The present invention discloses a method and system for accurately analyzing the phenotype of leopard gill perch based on deep learning. The key frame images of leopard gill perch are obtained through video monitoring, and the deep learning CNN model is used to identify fish schools, segment fish image areas and extract phenotypic features. A breeding cycle is selected, and the image area of ​​leopard gill perch is divided into sub-areas. The phenotypic features and preset variation feature data are calculated for feature differences to obtain variation characteristic values, and weights are assigned to each sub-area. The system stores the image and feature data of each identified target. In real-time fish school variation identification and statistical analysis, similarity calculations are performed between real-time targets and existing targets in the database to achieve repeatability detection and accurate target tracking of leopard gill perch. Through the present invention, rapid and accurate phenotypic analysis and target tracking of leopard gill perch can be achieved, and the accuracy of breeding analysis and breeding efficiency can be improved.

[0154] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0155] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0156] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0157] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0158] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0159] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for accurate phenotypic analysis of leopard gill perch based on deep learning, characterized in that: include: During multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set; A CNN image analysis model based on deep learning was constructed, and the key image set was imported into the image analysis model for fish school recognition, fish region segmentation, and fish phenotypic feature extraction. Feature extraction included contour texture features, color features, and texture distribution features, and the target image region and target feature data based on the leopard spiny perch image were obtained. A breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the feature difference calculation based on the standard Euclidean distance is performed on the current target feature data and the preset variation feature data. The difference calculation process is calculated based on each sub-area separately, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted, and the corresponding weight is assigned to each sub-area based on the variation characteristic value; In the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database; In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain the similarity index. The similarity index is used to perform repeatability detection on the real-time identification target and track the identification target; In each breeding cycle, the fish school information and mutation identification results are sent to the preset terminal device.

2. The method for accurately analyzing the phenotype of Leopard gill perch based on deep learning according to claim 1, characterized in that: In the above method, the target breeding area is monitored by a video monitoring device during multiple breeding cycles, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set, which is specifically: Set the breeding cycle, set up multiple monitoring points in the target breeding area, and set up video monitoring devices at each monitoring point; Use video monitoring devices to monitor the target breeding area and obtain video data of leopard gill perch; Key frames are extracted from video data to obtain key image sets, which are then preprocessed by image denoising, enhancement, and standardization.

3. The method for accurate phenotypic analysis of Leopard gill perch based on deep learning according to claim 1, characterized in that: The CNN image analysis model based on deep learning is constructed, and the key image set is imported into the image analysis model to perform fish school recognition, fish region segmentation and fish phenotypic feature extraction, and the feature extraction includes extraction of contour texture features, color features, and texture distribution features, and the target image region and target feature data based on the leopard gill perch image are obtained, specifically: The CNN image analysis model based on deep learning includes convolutional layer, pooling layer and fully connected layer; Based on historical image data, the image analysis model was trained to extract features and identify leopard gill perch; Import the key image set into the image analysis model for fish school recognition, perform fish school recognition through each image, and perform fish area segmentation on the leopard gill perch based on the recognition results to obtain the target image area; Image segmentation is performed based on the target image area and based on the image analysis model, the convolution layer and the pooling layer are used to perform convolution operations on the input image and extract local edge and texture features to obtain contour texture features; Performing color statistics on the target image area by means of a color histogram and generating corresponding color feature data based on the color statistical data in the histogram; The body color and pattern distribution is analyzed through contour texture features and color feature data to form texture distribution features based on body color and pattern.

4. The method for accurately analyzing the phenotype of Leopard gill perch based on deep learning according to claim 3 is characterized in that: The method selects a breeding cycle, obtains the current identified fish information, and marks the corresponding target image area and target feature data as the current target image area and the current target feature data, respectively. The current target image area is divided into multiple sub-areas, and the current target feature data and the preset variation feature data are subjected to feature difference calculation based on the standard Euclidean distance. The difference calculation process is based on each sub-area, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area. The variation characteristic values ​​are sorted and the corresponding weights are assigned to each sub-area based on the variation characteristic values. Specifically, A breeding cycle is selected, and a key image set is acquired and the identification and analysis of the leopard gill perch are performed on the breeding cycle to obtain current identification fish school information; By using the current identified fish school information, the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively; For the current target image area, each body part of the leopard gill perch is identified and segmented to obtain multiple sub-areas; Taking each sub-region as the analysis unit, extract the feature data of the sub-region from the current target feature data and convert it into a feature vector with the preset variation feature data, and calculate the feature difference based on the standard Euclidean distance to obtain the feature difference value of each sub-region; The variation characteristic value of each sub-region is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted and the corresponding weight is assigned to each sub-region based on the variation characteristic value.

5. The method for accurately analyzing the phenotype of Leopard gill perch based on deep learning according to claim 4, characterized in that: In the selected breeding cycle, based on the current identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database, specifically: According to the current fish school identification information, the target image area corresponding to each identification target is mapped and associated with the target feature data to form a data table and store it in the system database.

6. The method for accurate phenotypic analysis of Leopard gill perch based on deep learning according to claim 5, characterized in that: In each breeding cycle, fish school variation identification and statistics are performed through the target image area and the target feature data. In the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain a similarity index. The real-time identification target is detected for repeatability and the identification target is tracked and marked through the similarity index, specifically: In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked; Mark the target feature data of the real-time recognition target as the first feature data, and select a target feature data corresponding to the recognition target from the system database and mark it as the second feature data; The first feature data and the second feature data are similarly calculated. The calculation process is based on each sub-region. The similarity calculation is performed by converting the first feature data and the second feature data into feature vectors and calculating the similarity based on the standard Euclidean distance to obtain M similarities, where M is the number of sub-regions. Based on the corresponding weights, the M similarities are weighted averaged to obtain a similarity index; Determine whether the similarity index is greater than a preset similarity, and if so, determine that the real-time recognition target and the selected recognition target are the same target; Based on the similarity index, the real-time recognition target is detected for repeatability and the recognition target is tracked and marked.

7. The method for accurate phenotypic analysis of Leopard gill perch based on deep learning according to claim 6, characterized in that: In each breeding cycle, the fish school information and the variation identification result are sent to a preset terminal device, which includes a mobile terminal and a computer terminal.

8. A deep learning-based precise phenotypic analysis system for leopard gill perch, characterized in that: The system includes: a memory and a processor, wherein the memory includes a leopard gill perch phenotype precision analysis program based on deep learning, and the leopard gill perch phenotype precision analysis program based on deep learning is executed by the processor to implement the following steps: During multiple breeding cycles, the target breeding area is monitored by a video monitoring device, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set; A CNN image analysis model based on deep learning was constructed, and the key image set was imported into the image analysis model for fish school recognition, fish region segmentation, and fish phenotypic feature extraction. Feature extraction included contour texture features, color features, and texture distribution features, and the target image region and target feature data based on the leopard spiny perch image were obtained. A breeding cycle is selected, the current identified fish school information is obtained, and the corresponding target image area and target feature data are marked as the current target image area and the current target feature data respectively, the current target image area is divided into multiple sub-areas, and the feature difference calculation based on the standard Euclidean distance is performed on the current target feature data and the preset variation feature data. The difference calculation process is calculated based on each sub-area separately, and the variation characteristic value of each sub-area is calculated in combination with the number of identified targets in the target breeding area, the variation characteristic values ​​are sorted, and the corresponding weight is assigned to each sub-area based on the variation characteristic value; In the selected breeding cycle, based on the currently identified fish school information, the target image area and target feature data corresponding to each identified target are stored in the system database; In each breeding cycle, fish school variation identification and statistics are performed through the target image area and target feature data. During the statistical process, a real-time identification target is marked, and similarity calculation is performed based on the target feature data of the real-time identification target and the existing target feature data in the system database. The similarity calculation is combined with the similarity and weight of each sub-area for comprehensive calculation to obtain the similarity index. The similarity index is used to perform repeatability detection on the real-time identification target and track the identification target; In each breeding cycle, the fish school information and mutation identification results are sent to the preset terminal device.

9. The deep learning-based precise phenotype analysis system for leopard gill perch according to claim 8, characterized in that: In the above method, the target breeding area is monitored by a video monitoring device during multiple breeding cycles, and video data of leopard gill perch is obtained, and key frames are extracted from the video data to obtain a key image set, which is specifically: Set the breeding cycle, set up multiple monitoring points in the target breeding area, and set up video monitoring devices at each monitoring point; Use video monitoring devices to monitor the target breeding area and obtain video data of leopard gill perch; Key frames are extracted from video data to obtain key image sets, which are then preprocessed by image denoising, enhancement, and standardization.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a deep learning-based precise phenotypic analysis program for leopard gill perch. When the deep learning-based precise phenotypic analysis program for leopard gill perch is executed by a processor, the steps of the deep learning-based precise phenotypic analysis method for leopard gill perch as described in any one of claims 1 to 7 are implemented.

Images