An aquatic biodiversity monitoring system and method based on image recognition

Abstract

The present application provides an aquatic biodiversity monitoring system and method based on image recognition, which acquires underwater images of a monitoring water area, determines the substrate area without aquatic organisms and the potential biological areas to be verified in the underwater images; determines the color space distance between the potential biological areas and the substrate area, and determines the color texture characteristics of the texture structure of the potential biological areas based on the color space distance in combination with the gray-level co-occurrence matrix of the potential biological areas; extracts biological candidate areas according to the color texture characteristics and the contour characteristics of the potential biological areas, and determines the spatial distribution entropy of all texture breakpoints in the biological candidate areas; determines the mimicry labels of the mimic aquatic organisms in the biological candidate areas based on the spatial distribution entropy and texture structure of all texture breakpoints in the biological candidate areas, and generates a dynamic map for aquatic biodiversity monitoring according to all the mimicry labels. By adopting the solution of the present application, the distinction between mimic aquatic organisms and the water area substrate can be realized based on the extraction of appearance visual features.

Description

Technical Field

[0001] This application relates to the field of underwater image recognition technology. More specifically, this application relates to a monitoring system and method for aquatic biodiversity based on image recognition. Background Art

[0002] Underwater image recognition, as an important research direction in the cross - field of computer vision and marine science, with the help of high - resolution underwater camera equipment and deep - learning algorithms, enables researchers to efficiently and automatically identify and classify targets in the water environment, thereby improving the intelligent level of tasks such as water environment ecological monitoring and water resource assessment. In terms of aquatic biodiversity monitoring, underwater image recognition can make up for the deficiencies brought by the low efficiency and large sample interference of traditional observation methods, and provide data support for ecological protection and biological population research. With the introduction of methods such as edge computing, image enhancement, and multi - modal fusion, underwater image recognition is developing towards high precision, real - time performance, and environmental adaptability.

[0003] Underwater image recognition is mainly realized through processes such as image segmentation, feature extraction, and target detection. Typical methods include structure segmentation technology based on edge detection, feature extraction algorithms based on texture and color space, and underwater target recognition models trained using convolutional neural networks. These methods can relatively effectively cope with the interference of factors such as light changes, water turbidity, and complex backgrounds, improving the recognition accuracy and robustness of the target area. However, existing methods still face significant challenges when identifying mimic aquatic organisms. Mimic aquatic organisms simulate the surrounding bottom substrate environment by changing appearance features such as color, texture, and contour, making it difficult for traditional image recognition technology to accurately distinguish mimic organisms from the natural background, thus interfering with the accuracy of aquatic biodiversity monitoring. Therefore, how to distinguish mimic aquatic organisms from the water bottom substrate based on the extraction of appearance visual features has become a difficult problem in the industry. Summary of the Invention

[0004] This application provides a monitoring system and method for aquatic biodiversity based on image recognition, which can distinguish mimic aquatic organisms from the water bottom substrate based on the extraction of appearance visual features.

[0005] In a first aspect, this application provides an image recognition method for aquatic biodiversity monitoring, which is applied to an aquatic biodiversity monitoring system. The method includes the following steps:

[0006] Obtain an underwater image of the monitored water area, and determine the bottom substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image;

[0007] Determine the color space distance between each potential biological region and the substrate region respectively, and determine the color texture features of the texture structure of each potential biological region based on the color space distance and the gray-level co-occurrence matrix of each potential biological region;

[0008] Further extract a plurality of biological candidate regions from all potential biological regions according to the color texture features and the contour features of each potential biological region, and determine the spatial distribution entropy of all texture breakpoints in each biological candidate region;

[0009] Determine the mimicry label of the mimicry aquatic organisms in each biological candidate region based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate region, and generate a dynamic map for monitoring aquatic biodiversity according to all the mimicry labels.

[0010] In some embodiments, determining the substrate region without aquatic organisms and a plurality of potential biological regions to be verified in the underwater image specifically includes:

[0011] Extract a plurality of characteristic edge pixels of the substrate of the monitoring water area from the underwater image;

[0012] Determine the edge curvature of each characteristic edge pixel;

[0013] Divide the substrate region without aquatic organisms and a plurality of potential biological regions to be verified in the underwater image based on all the edge curvatures and the texture gradient field of the substrate of the underwater image.

[0014] In some embodiments, extracting a plurality of characteristic edge pixels of the substrate of the monitoring water area from the underwater image specifically includes:

[0015] Preprocess the underwater image to obtain a normalized image without color offset;

[0016] Determine the vertical texture gradient and horizontal texture gradient of each substrate pixel in the normalized image;

[0017] Determine the substrate characteristic value of each substrate pixel based on the vertical texture gradient and the horizontal texture gradient;

[0018] Extract a plurality of characteristic edge pixels of the substrate of the monitoring water area from all the substrate pixels according to the substrate characteristic value of each substrate pixel.

[0019] In some embodiments, determining the edge curvature of each characteristic edge pixel specifically includes:

[0020] Construct a plurality of texture edge segments of the substrate of the monitoring water area based on all the characteristic edge pixels;

[0021] Select a texture edge segment as the selected texture edge segment, and determine the adjacent pixel pairs of each feature edge pixel in the selected texture edge segment;

[0022] Determine the edge curvature of each feature edge pixel in the selected texture edge segment according to the coordinate information of the adjacent pixel pairs;

[0023] Continue to determine the edge curvature of each feature edge pixel in the remaining texture edge segments.

[0024] In some embodiments, dividing the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image based on all the edge curvatures and the texture gradient field of the substrate of the underwater image specifically includes:

[0025] Obtain the vertical texture gradient and horizontal texture gradient of each substrate pixel;

[0026] Construct the texture gradient field of the substrate of the underwater image based on the vertical texture gradient and horizontal texture gradient of each substrate pixel;

[0027] Obtain all the texture edge segments of the substrate of the monitoring water area, and determine the texture uniformity index of each texture edge segment based on the texture gradient field;

[0028] Divide the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the texture uniformity index of each texture edge segment and all the edge curvatures.

[0029] In some embodiments, dividing the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the texture uniformity index of each texture edge segment and all the edge curvatures specifically includes:

[0030] Select a texture edge segment, and extract all the feature edge pixels in the texture edge segment;

[0031] Obtain the edge curvature corresponding to each feature edge pixel;

[0032] Determine the edge label of the texture edge segment based on the texture uniformity index of the texture edge segment and all the edge curvatures corresponding to the texture edge segment;

[0033] Continue to determine the edge labels of the remaining texture edge segments;

[0034] Divide the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the edge labels of all the texture edge segments.

[0035] In some embodiments, determining the color texture features of the texture structure of each potential biological region based on the color space distance in combination with the gray level co-occurrence matrix of each potential biological region specifically includes:

[0036] Obtain the normalized image of the underwater image, and convert the normalized image into a grayscale processed image;

[0037] Intercept the grayscale sub-image corresponding to each potential biological region in the grayscale processed image;

[0038] Determine the gray level co-occurrence matrix of each grayscale sub-image;

[0039] Extract the texture feature parameters of each potential biological region from the gray level co-occurrence matrix;

[0040] Construct the color texture features of the texture structure of each potential biological region based on the texture feature parameters and the color space distance.

[0041] In some embodiments, further extracting a plurality of biological candidate regions from all potential biological regions according to the color texture features and the contour features of each potential biological region specifically includes:

[0042] Determine the contour features of each potential biological region;

[0043] Invoke a pre-trained classifier model;

[0044] Based on the classifier model, extract a plurality of biological candidate regions from all potential biological regions in combination with the color texture features and the contour features of each potential biological region.

[0045] In some embodiments, determining the spatial distribution entropy of all texture breakpoints in each biological candidate region specifically includes:

[0046] Select a biological candidate region as the selected biological candidate region, and obtain the gradient feature sequence of each pixel point in the selected biological candidate region;

[0047] Extract a plurality of texture breakpoints from all pixel points according to the gradient feature sequence of each pixel point;

[0048] Determine the spatial distribution entropy of all texture breakpoints in the selected biological candidate region;

[0049] Continue to determine the spatial distribution entropy of all texture breakpoints in the remaining biological candidate regions.

[0050] In a second aspect, the present application provides an aquatic biodiversity monitoring system based on image recognition, which includes an image recognition unit, and the image recognition unit includes:

[0051] An acquisition module, configured to acquire an underwater image of a monitored water area, and determine a substrate area without aquatic organisms and a plurality of potential biological areas to be verified in the underwater image;

[0052] A processing module, configured to respectively determine the color space distance between each potential biological area and the substrate area, and determine the color texture features of the texture structure of each potential biological area based on the color space distance in combination with the gray-level co-occurrence matrix of each potential biological area;

[0053] The processing module is further configured to further extract a plurality of biological candidate areas from all the potential biological areas according to the color texture features and the contour features of each potential biological area, and determine the spatial distribution entropy of all texture breakpoints in each biological candidate area;

[0054] An execution module, configured to determine the mimicry labels of mimic aquatic organisms in each biological candidate area based on the spatial distribution entropy and the texture structure of all texture breakpoints in each biological candidate area, and generate a dynamic map for monitoring aquatic biodiversity according to all the mimicry labels.

[0055] In a third aspect, the present application provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to obtain the code and execute the above-mentioned method for monitoring aquatic biodiversity based on image recognition.

[0056] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned method for monitoring aquatic biodiversity based on image recognition is implemented.

[0057] The technical solutions provided by the disclosed embodiments of the present application have the following beneficial effects:

[0058] In the system and method for monitoring aquatic biodiversity based on image recognition provided by the present application, first, an underwater image of a monitored water area is acquired, and a substrate area without aquatic organisms and a plurality of potential biological areas to be verified in the underwater image are determined; second, the color space distance between each potential biological area and the substrate area is respectively determined, and the color texture features of the texture structure of each potential biological area are determined based on the color space distance in combination with the gray-level co-occurrence matrix of each potential biological area; then, a plurality of biological candidate areas are further extracted from all the potential biological areas according to the color texture features and the contour features of each potential biological area, and the spatial distribution entropy of all texture breakpoints in each biological candidate area is determined; finally, the mimicry labels of mimic aquatic organisms in each biological candidate area are determined based on the spatial distribution entropy and the texture structure of all texture breakpoints in each biological candidate area, and a dynamic map for monitoring aquatic biodiversity is generated according to all the mimicry labels.

[0059] It can be seen that this application can distinguish mimetic aquatic organisms from water bottom sediments based on the extraction of appearance visual features. First, in underwater images, divide the bottom sediment area without aquatic organisms and the potential biological areas to be verified, and identify the surrounding bottom sediment environment and the image areas that may contain aquatic organisms, so as to facilitate subsequent area comparison, increase the efficiency of mimetic organism recognition, and then improve the accuracy of aquatic biodiversity monitoring. Second, based on the color space distance and the gray-level co-occurrence matrix of the potential biological areas, determine the color texture features of the texture structure of the potential biological areas, which reflect the overall color structure and texture organization complexity of the potential biological areas, so as to improve the description accuracy of the potential biological areas, thereby increasing the accuracy of mimetic organism recognition. Then, further extract the biological candidate areas from all potential biological areas according to the color texture features and the contour features of the potential biological areas to determine the real location where mimetic aquatic organisms exist, and track and describe them, so as to enhance the accuracy of aquatic biodiversity monitoring. Further, based on the spatial distribution entropy and texture structure of all texture breakpoints in the biological candidate areas, determine the mimetic labels of the mimetic aquatic organisms in the biological candidate areas to measure the mimetic degree of the mimetic organisms, which is conducive to the behavior analysis and tracking of the mimetic organisms, thereby increasing the dynamics and timeliness of diversity detection. Finally, generate a dynamic map of aquatic biodiversity monitoring according to the mimetic labels to dynamically display the structural evolution of aquatic biological communities in the time and space dimensions, so as to provide highly interpretable data support for the long-term monitoring, protection and management of water ecosystems. In summary, the technical solution provided by this application can distinguish mimetic aquatic organisms from water bottom sediments based on the extraction of appearance visual features. Description of the Drawings

[0060] Figure 1 is an exemplary flowchart of an aquatic biodiversity monitoring method based on image recognition according to some embodiments of the present application;

[0061] Figure 2 is an exemplary flowchart of determining color texture features according to some embodiments of the present application;

[0062] Figure 3 is an exemplary flowchart of determining mimetic labels according to some embodiments of the present application;

[0063] Figure 4 is a schematic structural diagram of an image recognition unit according to some embodiments of the present application;

[0064] Figure 5 is a schematic structural diagram of a computer device for implementing an aquatic biodiversity monitoring method based on image recognition according to some embodiments of the present application. Detailed Embodiments

[0065] To better understand the technical solution of this application, the technical solution of this application will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0066] Reference Figure 1 , this figure is an exemplary flowchart of a method for monitoring aquatic biodiversity based on image recognition shown according to some embodiments of this application. The method 100 for monitoring aquatic biodiversity based on image recognition mainly includes the following steps:

[0067] In step 101, an underwater image of the monitoring water area is acquired, and a substrate area without aquatic organisms and a plurality of potential biological areas to be verified in the underwater image are determined.

[0068] Specifically, when implementing, an underwater image of the monitoring water area is acquired, that is: an underwater image of the monitoring water area is acquired by deploying an underwater camera device with high-definition imaging capabilities in combination with an unmanned underwater shooting platform. Specifically, in this embodiment, an underwater camera equipped with a low-light enhancement sensor and a wide-angle lens is used, and the underwater camera is carried by a remotely operated underwater vehicle for shooting, so as to acquire an underwater image of the monitoring water area. Then, the underwater image is uploaded to a local storage device as the basic data source for subsequent aquatic organism image recognition. The underwater image represents the image data obtained by shooting with an underwater camera in the monitoring water area. By acquiring the underwater image, the visual information of the underwater environment and the target objects therein (for example, substrate and mimetic organisms) can be reflected.

[0069] In some embodiments, the substrate area without aquatic organisms and the plurality of potential biological areas to be verified in the underwater image can be determined in the following manner, that is:

[0070] Extract a plurality of characteristic edge pixels of the substrate of the monitoring water area from the underwater image;

[0071] Determine the edge curvature of each characteristic edge pixel;

[0072] Based on all the edge curvatures and the texture gradient field of the substrate of the underwater image, divide the substrate area without aquatic organisms and the plurality of potential biological areas to be verified in the underwater image.

[0073] Among them, in some embodiments, a plurality of characteristic edge pixels of the substrate of the monitoring water area can be extracted from the underwater image in the following manner, that is:

[0074] Preprocess the underwater image to obtain a standardized image without color offset;

[0075] Determine the vertical texture gradient and horizontal texture gradient of each substrate pixel in the standardized image;

[0076] Determine the substrate feature value of each substrate pixel based on the vertical texture gradient and the horizontal texture gradient;

[0077] Extract a plurality of characteristic edge pixels of the substrate of the monitored water area from all substrate pixels according to the substrate feature value of each substrate pixel.

[0078] When specifically implemented, preprocess the underwater image to obtain a standardized image without color offset, that is: the underwater image often has problems of uneven illumination and color offset. Therefore, it is necessary to preprocess the underwater image through the image processing software OpenCV (Open Source Computer Vision Library, OpenCV) to obtain a standardized image without color offset. Specifically, the preprocessing process in this embodiment includes denoising, contrast enhancement, and color correction. In addition, in other embodiments, other image processing software can also be used for processing, which is not limited here.

[0079] When specifically implemented, determine the vertical texture gradient and the horizontal texture gradient of each substrate pixel in the standardized image, that is: before identifying the mimicry organisms, it is considered that the pixel points in the image all reflect the visual information of the underwater substrate and its texture structure. Therefore, the pixel points in the standardized image are regarded as substrate pixels, and then the sobel gradient operator is used to calculate the vertical texture gradient and the horizontal texture gradient of each substrate pixel. The vertical texture gradient and the horizontal texture gradient represent the quantization values of the local changes of the substrate texture color. In addition, in other embodiments, other gradient operators can also be used for gradient calculation, which is not limited here.

[0080] When specifically implemented, determine the substrate feature value of each substrate pixel based on the vertical texture gradient and the horizontal texture gradient, that is: determine the gradient amplitude of each substrate pixel based on the vertical texture gradient and the horizontal texture gradient, and use the gradient amplitude as the substrate feature value of each substrate pixel. Specifically, directly substitute the vertical texture gradient and the horizontal texture gradient into the gradient amplitude calculation formula to obtain the gradient amplitude. The substrate feature value represents the measure of the gradient size, and the color change amplitude of the substrate pixels in the image can be reflected through the substrate feature value.

[0081] In specific implementation, a plurality of characteristic edge pixels of the substrate in the monitored water area are extracted from all substrate pixels according to the substrate characteristic values of each substrate pixel, that is: a characteristic threshold is preset in combination with historical experimental experience, the substrate characteristic values of all substrate pixels are compared with the characteristic threshold, and the substrate pixels with substrate characteristic values greater than the characteristic threshold are extracted. The extracted substrate pixels are used as characteristic edge pixels, and then a plurality of characteristic edge pixels of the substrate in the monitored water area are obtained. The characteristic edge pixels represent pixel points at the edge position in the image. These characteristic edge pixels are located in an area with a large color change amplitude, and this area is usually the edge position of the recognition target in the image, and further monitoring and verification are required for it.

[0082] Among them, in some embodiments, the edge curvature of each characteristic edge pixel can be determined by the following method, that is:

[0083] Construct a plurality of texture edge segments of the substrate in the monitored water area based on all the characteristic edge pixels;

[0084] Select a texture edge segment as the selected texture edge segment, and determine the adjacent pixel pairs of each characteristic edge pixel in the selected texture edge segment;

[0085] Determine the edge curvature of each characteristic edge pixel in the selected texture edge segment according to the coordinate information of the adjacent pixel pairs;

[0086] Continue to determine the edge curvature of each characteristic edge pixel in the remaining texture edge segments.

[0087] In specific implementation, a plurality of texture edge segments of the substrate in the monitored water area are constructed based on all the characteristic edge pixels, that is: edge tracking is performed on all the characteristic edge pixels to obtain a plurality of texture edge segments of the substrate in the monitored water area. Specifically, in this embodiment, the texture edge segments are constructed by the method of sequentially tracking points in the 8-neighborhood direction of the characteristic edge pixels to construct line segments. During the edge tracking process, the starting point and ending point of the edge segment of the texture edge segment also need to be marked, and the texture edge segment is output in the form of a point sequence for subsequent processing. The remaining specific edge construction process will not be elaborated here. In addition, in other embodiments, other methods can also be used for edge construction, such as the findContours function, connected component analysis, etc., which are not limited here. The texture edge segments represent the edge line segments in the image caused by texture changes, reflecting the organizational structure of the image texture pattern. By extracting the texture edge segments, it is beneficial to analyze the surface differences under non-significant shape boundaries, thereby providing data support for the distinction between mimetic organisms and substrate areas.

[0088] In specific implementation, adjacent pixel pairs of each feature edge pixel in the selected texture edge segment are determined, that is: the starting point and the ending point of the edge segment of the selected texture edge segment are obtained, and the point sequence corresponding to the selected texture edge segment is traversed from the starting point to the ending point of the edge segment, so as to obtain the adjacent pixel pairs of each feature edge pixel in the selected texture edge segment.

[0089] In specific implementation, the edge curvature of each feature edge pixel in the selected texture edge segment is determined according to the coordinate information of the adjacent pixel pairs, that is: a feature edge pixel is selected, and the coordinate information of the feature edge pixel and its corresponding adjacent pixel pairs is obtained through Open CV, and the edge curvature of the feature edge pixel is calculated based on the coordinate information in combination with the three-point method, so as to obtain the edge curvature of each feature edge pixel in the selected texture edge segment. In addition, in other embodiments, other methods can also be used to calculate the edge curvature, which is not limited here. The edge curvature represents the degree of bending of the texture edge segment at the feature edge pixel. The larger the edge curvature, the greater the degree of bending of the edge segment, and the greater the probability of representing the edge of rock / gravel. The smaller the edge curvature, the smaller the degree of bending of the edge segment, and the greater the probability of representing the contour of an animal. Determining the edge curvature can be used as a shape description feature for classifying the substrate area.

[0090] Among them, in some embodiments, the substrate area without aquatic organisms and multiple potential biological areas to be verified can be divided in the underwater image based on all the edge curvatures and the texture gradient field of the substrate of the underwater image by the following method, that is:

[0091] Obtain the vertical texture gradient and the horizontal texture gradient of each substrate pixel;

[0092] Construct the texture gradient field of the substrate of the underwater image based on the vertical texture gradient and the horizontal texture gradient of each substrate pixel;

[0093] Obtain all texture edge segments of the substrate of the monitoring water area, and determine the texture uniformity index of each texture edge segment based on the texture gradient field;

[0094] Divide the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the texture uniformity index of each texture edge segment and all the edge curvatures.

[0095] In specific implementation, a texture gradient field of the underwater image substrate is constructed based on the vertical texture gradient and the horizontal texture gradient of each substrate pixel, that is: select a substrate pixel, calculate the quotient of the vertical texture gradient and the horizontal texture gradient of this substrate pixel, substitute the quotient result into the arctangent function to obtain the gradient direction angle of this substrate pixel, and obtain the gradient amplitude of this substrate pixel. Combine the coordinate information, gradient amplitude, and gradient direction angle of this substrate pixel into a gradient feature sequence, and then obtain the gradient feature sequences corresponding to all substrate pixels. Combine all the gradient feature sequences according to the corresponding positions of the coordinate information to form the texture gradient field of the underwater image substrate. The texture gradient field represents a data field that describes the distribution pattern and directional characteristics of the substrate texture in the image. By constructing the texture gradient field, it is beneficial to perform substrate classification, thereby improving the accuracy of mimetic organism identification.

[0096] In specific implementation, a texture uniformity index of each texture edge segment is determined based on the texture gradient field, that is: select a texture edge segment, extract the gradient feature sequences corresponding to all feature edge pixels on this texture edge segment from the texture gradient field, and calculate the standard deviation of the direction angle and the coefficient of variation of the amplitude of this texture edge segment based on all the gradient feature sequences through the numpy component in Python. Take the mean of the standard deviation of the direction angle and the coefficient of variation of the amplitude as the texture uniformity index of this texture edge segment, and then obtain the texture uniformity indexes of each texture edge segment. The specific calculation process is not elaborated here. The texture uniformity index represents an index for measuring whether the texture edge segment is located in the texture transition area. By determining the texture uniformity index, the consistency of the texture around the edge segment can be reflected.

[0097] Among them, in some embodiments, the substrate area without aquatic organisms and multiple potential biological areas to be verified can be divided in the underwater image according to the texture uniformity index of each texture edge segment and all the edge curvatures in the following manner, that is:

[0098] Select a texture edge segment and extract all the feature edge pixels in this texture edge segment;

[0099] Obtain the edge curvature corresponding to each feature edge pixel;

[0100] Determine the edge label of this texture edge segment based on the texture uniformity index of this texture edge segment and all the edge curvatures corresponding to this texture edge segment;

[0101] Continue to determine the edge labels of the remaining texture edge segments;

[0102] Divide the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the edge labels of all texture edge segments.

[0103] In specific implementation, based on the texture uniformity index of the texture edge segment and all the edge curvatures corresponding to the texture edge segment, determine the edge label of the texture edge segment, that is: obtain the curvature mean value of all the edge curvatures corresponding to the texture edge segment, and then, based on the texture uniformity index and the curvature mean value of the texture edge segment, combine the double-threshold method to assign an edge label to the texture edge segment. Specifically, set a uniformity threshold and a curvature threshold in combination with historical experimental experience. When the texture uniformity index is less than the uniformity threshold and the curvature mean value is greater than the curvature threshold, the corresponding texture edge segment is used as the substrate edge; otherwise, the corresponding texture edge segment is used as a potential biological edge. Take the above determination result as the edge label of the texture edge segment. By determining the edge label, the classification of the edge segment can be justified.

[0104] In specific implementation, divide the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the edge labels of all the texture edge segments, that is: connect the texture edge segments of the same type through an edge connection algorithm based on region growing to form multiple closed regions. The closed region formed by connecting the substrate edges is used as the substrate area, and the remaining closed regions are used as potential biological areas. Then, divide the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image.

[0105] It should be noted that in this application, the substrate area refers to the area in the image that is confirmed not to contain aquatic organisms, and the surface of this area is mainly bare substrate, such as sandy land, muddy land, rocks, shell fragments, etc. In addition, in this application, the potential biological area refers to the image area that may contain aquatic organisms detected by preliminary image processing means, but it has not been finally confirmed whether it is a biological organism and needs to be further verified. By dividing the substrate area and the potential biological area, it is beneficial to perform subsequent area comparison, increase the efficiency of mimicry organism recognition, and then improve the accuracy of aquatic biodiversity monitoring.

[0106] In step 102, respectively determine the color space distance between each potential biological area and the substrate area, and based on the color space distance, combine the gray-level co-occurrence matrix of each potential biological area to determine the color texture feature of the texture structure of each potential biological area.

[0107] In some embodiments, the color space distance between each potential biological area and the substrate area can be determined by the following method, that is:

[0108] Select a potential biological area as the selected potential biological area;

[0109] Perform color space transformation on the selected potential biological area to obtain the color attribute information of each pixel point in the selected potential biological area;

[0110] Determining a first color representation value for each pixel in the selected potential biological area based on the color attribute information;

[0111] Determining a second color representation value for each pixel in the substrate area;

[0112] Determining a color space distance between the selected potential biological region and the substrate region based on a first color representation value of each pixel point in the selected potential biological region and a second color representation value of each pixel point in the substrate region;

[0113] Continue to determine the color space distance between the remaining potential biological areas and the substrate area.

[0114] In the specific implementation, the color space conversion is performed on the selected potential biological area to obtain the color attribute information of each pixel in the selected potential biological area, that is: the selected potential biological area is converted from RGB color space to Lab color space through Open CV, and then the L value, a value, and b value of each pixel in the selected potential biological area are obtained. The L value, a value, and b value are combined into color attribute information, thereby obtaining the color attribute information of each pixel in the selected potential biological area.

[0115] In specific implementation, the first color representation value of each pixel point in the selected potential biological area is determined based on the color attribute information, that is: a pixel point in the selected potential biological area is selected, and the average of the L value, a value, and b value in the color attribute information corresponding to the pixel point is used as the first color representation value of the pixel point, thereby obtaining the first color representation value of each pixel point in the selected potential biological area. The first color representation value represents the average expression of the brightness and color tendency of the pixel point in the image area. By determining the first color representation value, subsequent color texture feature extraction and comparative analysis can be facilitated.

[0116] In specific implementation, the second color representation value of each pixel point in the substrate area is determined, that is, the substrate area is processed using the same method as above, and the average of the L value, a value, and b value of each pixel point in the substrate area is used as the second color representation value of each pixel point in the substrate area.

[0117] In specific implementation, the color space distance between the selected potential biological area and the substrate area is determined based on the first color representation value of each pixel point in the selected potential biological area and the second color representation value of each pixel point in the substrate area, that is, the mean of the first color representation value of all pixels in the selected potential biological area and the mean of the second color representation value of all pixels in the substrate area are calculated respectively. By taking the mean, the average Lab color perception of the area can be represented, and then the Euclidean distance of the two means is taken as the color space distance between the selected potential biological area and the substrate area.

[0118] It should be noted that in this application, the color space distance represents the degree of color difference between two regions. The larger the color space distance, the more obvious the color difference between the two regions. Conversely, the closer the colors of the two regions are. By determining the color space distance, it is beneficial to the subsequent extraction of color texture features and the optimization of regional comparison work, thereby improving the efficiency of mimicry biometric recognition and enhancing the reliability of aquatic biodiversity monitoring.

[0119] In some embodiments, based on the color space distance and combined with the gray-level co-occurrence matrix of each potential biological region, the color texture features of the texture structure of each potential biological region can be determined in the following manner, that is:

[0120] Obtain the standardized image of the underwater image and convert the standardized image into a grayscale processed image;

[0121] Intercept the grayscale sub-image corresponding to each potential biological region in the grayscale processed image;

[0122] Determine the gray-level co-occurrence matrix of each grayscale sub-image;

[0123] Extract the texture feature parameters of each potential biological region from the gray-level co-occurrence matrix;

[0124] Construct the color texture features of the texture structure of each potential biological region based on the texture feature parameters and the color space distance.

[0125] When specifically implemented, the standardized image is converted into a grayscale processed image through OpenCV. The specific process is not elaborated here. In addition, in other embodiments, other methods can also be used for grayscale conversion, which is not limited here.

[0126] When specifically implemented, the grayscale sub-image corresponding to each potential biological region is intercepted in the grayscale processed image through region masking operation. The region masking operation is an existing technology in image processing. In this embodiment, only its call is made, so the specific implementation process is not elaborated here. In addition, in other embodiments, other methods can also be used for intercepting grayscale sub-images, which is not limited here.

[0127] When specifically implemented, the gray-level co-occurrence matrix of each grayscale sub-image is determined, that is: the gray-level co-occurrence matrix of each grayscale sub-image is determined through Python. Specifically, in this embodiment, the skimage.feature.greycomatrix() function is called through Python to calculate the gray-level co-occurrence matrix, and then the gray-level co-occurrence matrix of each grayscale sub-image is determined. In addition, in other embodiments, other methods can also be used to determine the gray-level co-occurrence matrix, which is not limited here.

[0128] In specific implementation, texture feature parameters of each potential biological region are extracted from the gray-level co-occurrence matrix, that is: contrast, energy, entropy, and homogeneity of the potential biological region are extracted from the gray-level co-occurrence matrix, and the parameters obtained by the above extraction are used as the texture feature parameters of the potential biological region, so as to obtain the texture feature parameters of each potential biological region. The texture feature parameters refer to the quantitative representation of the spatial relationship and repeated pattern of pixel gray levels in the image. By extracting the texture feature parameters, the roughness, complexity, directionality, and uniformity of the image region can be reflected, so as to quantitatively describe the texture structure and spatial distribution law of the image region.

[0129] In specific implementation, color texture features of the texture structure of each potential biological region are constructed based on the texture feature parameters and the color space distance, that is: the texture feature parameters and the color space distance are fused into a composite feature vector, and the composite feature vector is used as the color texture feature of the texture structure of each potential biological region.

[0130] It should be noted that in this application, the color texture feature represents the comprehensive expression of the texture structure and color of the potential biological region. Refer to Figure 2 As shown, this figure is an exemplary flowchart for determining the color texture feature according to some embodiments of this application. The color texture feature not only reflects the overall color structure of the potential biological region, but also describes the texture organization complexity of the potential biological region. By determining the color texture feature, the description accuracy of the potential biological region can be improved, thereby increasing the accuracy of mimicry biometric recognition, and further improving the efficiency and reliability of diversity monitoring.

[0131] In step 103, based on the color texture feature and the contour feature of each potential biological region, multiple biological candidate regions are further extracted from all potential biological regions, and the spatial distribution entropy of all texture breakpoints in each biological candidate region is determined.

[0132] In some embodiments, the further extraction of multiple biological candidate regions from all potential biological regions based on the color texture feature and the contour feature of each potential biological region can be implemented in the following manner, that is:

[0133] Determine the contour feature of each potential biological region;

[0134] Call the pre-trained classifier model;

[0135] Based on the classifier model, multiple biological candidate regions are extracted from all potential biological regions by combining the color texture feature and the contour feature of each potential biological region.

[0136] In specific implementation, the contour features of each potential biological region are determined, that is: the contour smoothness of each potential biological region is determined by Fourier descriptors, and the obtained contour smoothness is used as the contour feature of each potential biological region. Specifically, the Fourier descriptors are used to represent the region contours of each potential biological region in the frequency domain, and the proportion of high-frequency components after the frequency domain representation is evaluated. The low-frequency components describe the general shape of the region contour, and the high-frequency components describe the details, burrs, and noises of the region contour edges. Therefore, the proportion of high-frequency components is used as the contour smoothness, and then the contour features of each potential biological region are obtained. In addition, in other embodiments, other methods can also be used to determine the contour features, such as boundary coherence scoring, etc., which are not limited here.

[0137] It should be noted that in this application, the contour feature is a comprehensive measurement index representing the complexity and smoothness of the region contour. The larger the contour feature, the more uneven the contour of the potential biological region, and the greater the probability of representing the substrate (gravel, sand), and vice versa, the greater the probability of representing aquatic organisms. By determining the contour feature to judge the smoothness and complexity of the contour, further data support is provided for the identification and judgment of mimetic organisms.

[0138] In specific implementation, a pre-trained classifier model is called, that is: a classifier model pre-trained with a large number of samples of real mimetic aquatic biological regions and non-biological substrate regions is called. Specifically, in this embodiment, a support vector machines (SVM) classifier model is called. The SVM model can identify whether the monitoring object is a mimetic biological region. In addition, in other embodiments, other classifier models can also be called, such as a random forest model, etc., which are not limited here.

[0139] In specific implementation, based on the classifier model, multiple biological candidate regions are extracted from all potential biological regions by combining the color texture features and contour features of each potential biological region, that is: the color texture features and contour features of each potential biological region are used as the input features of the classifier model. After being recognized by the classifier model, it can be recognized whether the potential biological region corresponding to the input features is a mimetic biological region and the biological species can be recognized, and then multiple biological candidate regions are extracted. The biological candidate regions represent the real positions where mimetic aquatic organisms exist. However, due to the differences in the mimetic degree, the extracted biological candidate regions need to be further described to enhance the accuracy of aquatic biodiversity monitoring.

[0140] In some embodiments, the spatial distribution entropy of all texture breakpoints in each biological candidate region can be determined in the following manner, that is:

[0141] Select a biological candidate area as the selected biological candidate area, and obtain the gradient feature sequence of each pixel point in the selected biological candidate area;

[0142] Extract a plurality of texture breakpoints from all pixel points according to the gradient feature sequence of each pixel point;

[0143] Determine the spatial distribution entropy of all texture breakpoints in the selected biological candidate area;

[0144] Continue to determine the spatial distribution entropy of all texture breakpoints in the remaining biological candidate areas.

[0145] When specifically implemented, a plurality of texture breakpoints are extracted from all pixel points according to the gradient feature sequence of each pixel point, that is: first, set a direction angle threshold in combination with historical experimental experience; then extract the gradient direction angle of each pixel point from the gradient feature sequence, compare all the gradient direction angles with the direction angle threshold, extract the pixel points corresponding to all the gradient direction angles greater than the direction angle threshold, and use the extracted pixel points as texture breakpoints, thereby obtaining a plurality of texture breakpoints.

[0146] It should be noted that in this application, the texture breakpoint represents the position of texture mutation and the edge break point in the image area. By extracting the texture breakpoint, it is beneficial to enrich the description of the appearance texture characteristics of the mimetic organism, thereby improving the accuracy of diversity monitoring.

[0147] Among them, in some embodiments, the spatial distribution entropy of all texture breakpoints in the selected biological candidate area can be determined by the following method, that is:

[0148] Divide the selected biological candidate area into a plurality of grids with uniform size;

[0149] Count the number of texture breakpoints in each grid;

[0150] Respectively, take the quotient of the number of texture breakpoints in each grid and the total number of all texture breakpoints as the spatial distribution probability of the texture breakpoints in each grid;

[0151] Based on the spatial distribution probability of the texture breakpoints in each grid, combine with the Shannon entropy formula to determine the spatial distribution entropy of all texture breakpoints in the selected biological candidate area.

[0152] It should be noted that in this application, the spatial distribution entropy represents an index for measuring the uniformity and complexity of the spatial coordinate distribution of texture breakpoints in the biological candidate area. By determining the spatial distribution entropy, the appearance visual characteristics of the mimetic organism in the biological candidate area during camouflage can be further described, thereby improving the credibility of aquatic biodiversity monitoring.

[0153] In step 104, based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate region, determine the mimicry label of the mimic aquatic organisms in each biological candidate region, and generate a dynamic map for monitoring aquatic biodiversity according to all the mimicry labels.

[0154] In some embodiments, refer to Figure 3 As shown, this figure is an exemplary flowchart for determining the mimicry label according to some embodiments of the present application. In this embodiment, the following steps can be used to determine the mimicry label of the mimic aquatic organisms in each biological candidate region based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate region:

[0155] First, in step 1041, set a detection window for detecting the difference in texture structure in each biological candidate region;

[0156] Secondly, in step 1042, obtain the gray-level co-occurrence matrix corresponding to each biological candidate region;

[0157] Then, in step 1043, determine the local variation characteristics of the texture structure in each biological candidate region according to the detection window and the gray-level co-occurrence matrix;

[0158] Furthermore, in step 1044, determine the mimicry score of the mimic organisms in each biological candidate region based on the spatial distribution entropy and the local variation characteristics of the texture structure of all texture breakpoints in each biological candidate region;

[0159] Finally, in step 1045, determine the mimicry label of the mimic aquatic organisms in each biological candidate region according to the mimicry score of the mimic organisms in each biological candidate region.

[0160] It should be noted that in this embodiment, the size of the detection window is selected as 5*5 pixel units. In addition, in other embodiments, the size of the detection window can be set according to specific application requirements, which is not limited here.

[0161] When specifically implemented, determine the local variation characteristics of the texture structure in each biological candidate region according to the detection window and the gray-level co-occurrence matrix, that is: select a biological candidate region, slide and traverse the gray-level co-occurrence matrix corresponding to the biological candidate region through the detection window corresponding to the biological candidate region, record the contrast in each sliding detection area, and then calculate the variance of all contrasts as the local variation characteristics of the texture structure in the biological candidate region, and then obtain the local variation characteristics of the texture structure in each biological candidate region. The local variation characteristics represent an index for measuring the local texture inconsistency in the biological candidate region. By determining the local variation characteristics, the mimicry ability of the mimic organisms can be reflected.

[0162] In specific implementation, the mimicry score of the mimicking organism in each biological candidate region is determined based on the spatial distribution entropy of all texture breakpoints and the local variation characteristics of the texture structure in each biological candidate region, that is: select a biological candidate region, calculate the mean value of the corresponding spatial distribution entropy and local variation characteristics of this biological candidate region, and use the calculated mean value as the mimicry score of the mimicking organism in this biological candidate region, thereby obtaining the mimicry scores of the mimicking organisms in each biological candidate region. The mimicry score is a comprehensive measurement index for the mimicry ability of the mimicking organism in the region. It should be noted that before calculating the mean value, the spatial distribution entropy and local variation characteristics need to be normalized to ensure the consistency of the data scale.

[0163] In specific implementation, the mimicry label of the mimicking aquatic organism in each biological candidate region is determined based on the mimicry score of the mimicking organism in each biological candidate region, that is: set two score thresholds according to historical practice experience, denoted as the first score threshold and the second score threshold (the first score threshold is less than the second score threshold), compare the mimicry score of the mimicking organism in each biological candidate region with the first score threshold and the second score threshold. When the mimicry score is less than the first score threshold, it is determined that the mimicking organism in the corresponding biological candidate region is a low-degree mimicking organism. When the mimicry score is between the first score threshold and the second score threshold, it is determined that the mimicking organism in the corresponding biological candidate region is a medium-degree mimicking organism. When the mimicry score is greater than the second score threshold, it is determined that the mimicking organism in the corresponding biological candidate region is a high-degree mimicking organism. Then, the determination result is used as the mimicry label of the mimicking aquatic organism in each biological candidate region. The mimicry label can measure the mimicry degree of the mimicking organism. By determining the mimicry label, the additional attributes of the biological candidate region can be increased, which is conducive to the behavioral analysis and tracking of the mimicking organism, and further increases the dynamics and timeliness of diversity detection.

[0164] In specific implementation, a dynamic map for monitoring aquatic organism diversity is generated based on all the mimicry labels, that is: first, the mimicking organisms in the biological candidate region are tracked through an object tracking algorithm based on deep learning to obtain the time-series data for monitoring aquatic organism diversity; then, the time-series data and all the mimicry labels are coordinate-bound with the geographic information system to generate a dynamic map including the species distribution density of the mimicking organisms and the spatial aggregation heat map of the mimicking organisms, and the biological contours of the mimicking organisms are real-time marked through an augmented reality device. The dynamic map can dynamically display the structural evolution of the aquatic organism community in the time and space dimensions, reflect the time-series changes in the distribution positions of the mimicking organisms, and also reveal the ecological diversity and camouflage strategies of the mimicking organisms, thereby providing strongly interpretable data support for the long-term monitoring, protection, and management of the water ecosystem.

[0165] In addition, on the other hand of the present application, in some embodiments, the present application provides an aquatic biodiversity monitoring system based on image recognition. The system includes an image recognition unit. Refer to Figure 4 , which is a schematic structural diagram of the image recognition unit shown in some embodiments of the present application. The image recognition unit 200 includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described as follows:

[0166] The acquisition module 201 is mainly used in the present application to acquire underwater images of the monitored water area, and determine the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater images;

[0167] The processing module 202 is mainly used in the present application to respectively determine the color space distance between each potential biological area and the substrate area, and determine the color texture features of the texture structure of each potential biological area based on the color space distance in combination with the gray-level co-occurrence matrix of each potential biological area;

[0168] The processing module 202 is further configured to further extract multiple biological candidate areas from all potential biological areas according to the color texture features and the contour features of each potential biological area, and determine the spatial distribution entropy of all texture breakpoints in each biological candidate area;

[0169] The execution module 203 is mainly used in the present application to determine the mimicry labels of the mimic aquatic organisms in each biological candidate area based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate area, and generate a dynamic map for aquatic biodiversity monitoring according to all the mimicry labels.

[0170] In addition, the present application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to obtain the code and execute the above-mentioned aquatic biodiversity monitoring method based on image recognition.

[0171] In some embodiments, refer to Figure 5 , which is a schematic structural diagram of a computer device for implementing the aquatic biodiversity monitoring method based on image recognition shown in some embodiments of the present application. The above-mentioned aquatic biodiversity monitoring method based on image recognition in the embodiments can be implemented by Figure 5 the computer device shown. The computer device 300 includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0172] The processor 301 can be a general-purpose central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more for controlling the execution of the method for monitoring aquatic biodiversity based on image recognition in this application.

[0173] The communication bus 302 can be used to transfer information between the above components.

[0174] The memory 303 can be a read-only memory (ROM), or other types of static storage devices that can store static information and instructions, a random access memory (RAM), or other types of dynamic storage devices that can store information and instructions. It can also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disks, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to this. The memory 303 can exist independently and be connected to the processor 301 through the communication bus 302. The memory 303 can also be integrated with the processor 301.

[0175] Among them, the memory 303 is used to store the program code for executing the solution of this application, and is controlled by the processor 301 to execute. The processor 301 is used to execute the program code stored in the memory 303. The program code can include one or more software modules. The determination of the method for monitoring aquatic biodiversity based on image recognition in the above embodiments can be implemented by one or more software modules in the program code of the processor 301 and the memory 303.

[0176] The communication interface 304, using any device such as a transceiver, is used to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0177] In a specific implementation, as an example, a computer device may include multiple processors, and each of these processors can be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. The processor here can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0178] The above computer device can be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device can be a desktop computer, a laptop computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of the computer device.

[0179] In addition, the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above method for monitoring aquatic biodiversity based on image recognition is implemented.

[0180] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0181] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these changes and modifications.

Claims

1. An image recognition method for aquatic biodiversity monitoring, which is applied to an aquatic biodiversity monitoring system, and is characterized in that The method includes the following steps: Obtain an underwater image of the monitored water area, and determine a substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image; Respectively determine the color space distance between each potential biological area and the substrate area, and determine the color texture characteristics of the texture structure of each potential biological area based on the color space distance and the gray-level co-occurrence matrix of each potential biological area; Further extract multiple biological candidate areas from all potential biological areas according to the color texture characteristics and the contour characteristics of each potential biological area, and determine the spatial distribution entropy of all texture breakpoints in each biological candidate area; Determine the mimicry label of the mimic aquatic organisms in each biological candidate area based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate area, and generate a dynamic map for monitoring aquatic biodiversity according to all the mimicry labels; Among them, determining the mimicry label of the mimic aquatic organisms in each biological candidate area based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate area is realized by the following steps: Set a detection window for detecting the difference in texture structure in each biological candidate area; Obtain the gray-level co-occurrence matrix corresponding to each biological candidate area; Determine the local variation characteristics of the texture structure in each biological candidate area according to the detection window and the gray-level co-occurrence matrix; Determine the mimicry score of the mimic organisms in each biological candidate area based on the spatial distribution entropy and the local variation characteristics of the texture structure of all texture breakpoints in each biological candidate area; Determine the mimicry label of the mimic aquatic organisms in each biological candidate area according to the mimicry score of the mimic organisms in each biological candidate area; Among them, generating a dynamic map for monitoring aquatic biodiversity according to all the mimicry labels, that is: tracking the mimic organisms in the biological candidate area through an object tracking algorithm based on deep learning to obtain the time-series data for monitoring aquatic biodiversity; binding the time-series data and all the mimicry labels to coordinates with a geographic information system to generate a dynamic map including the distribution density of mimic biological species and the heat map of spatial aggregation of mimic organisms.

2. The method according to claim 1, characterized in that Determining the substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image specifically includes: Extract multiple characteristic edge pixels of the substrate of the monitored water area from the underwater image; Determine the edge curvature of each characteristic edge pixel; Divide a substrate area without aquatic organisms and multiple potential biological areas to be verified in the underwater image based on all the edge curvatures and the texture gradient field of the substrate of the underwater image.

3. The method according to claim 2, wherein Extracting multiple characteristic edge pixels of the substrate of the monitored water area from the underwater image specifically includes: Preprocess the underwater image to obtain a standardized image without color offset; Determine the vertical texture gradient and horizontal texture gradient of each substrate pixel in the standardized image; Determine the substrate characteristic value of each substrate pixel based on the vertical texture gradient and the horizontal texture gradient; Extract multiple characteristic edge pixels of the substrate of the monitored water area from all the substrate pixels according to the substrate characteristic value of each substrate pixel.

4. The method according to claim 2, wherein Determining the edge curvature of each characteristic edge pixel specifically includes: Construct a plurality of texture edge segments for monitoring the bottom sediment of the water area based on all feature edge pixels; Select one texture edge segment as the selected texture edge segment, and determine the adjacent pixel pairs of each feature edge pixel in the selected texture edge segment; Determine the edge curvature of each feature edge pixel in the selected texture edge segment according to the coordinate information of the adjacent pixel pairs; Continue to determine the edge curvature of each feature edge pixel in the remaining texture edge segments.

5. The method according to claim 2, wherein Dividing the bottom sediment area without aquatic organisms and multiple potential biological areas to be verified in the underwater image based on all the edge curvatures and the texture gradient field of the bottom sediment of the underwater image specifically includes: Obtain the vertical texture gradient and horizontal texture gradient of each bottom sediment pixel; Construct the texture gradient field of the bottom sediment of the underwater image based on the vertical texture gradient and horizontal texture gradient of each bottom sediment pixel; Obtain all texture edge segments of the bottom sediment of the monitoring water area, and determine the texture uniformity index of each texture edge segment based on the texture gradient field; Divide the bottom sediment area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the texture uniformity index of each texture edge segment and all the edge curvatures.

6. The method according to claim 5, wherein Dividing the bottom sediment area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the texture uniformity index of each texture edge segment and all the edge curvatures specifically includes: Select one texture edge segment, and extract all feature edge pixels in the texture edge segment; Obtain the edge curvature corresponding to each feature edge pixel; Determine the edge label of the texture edge segment based on the texture uniformity index of the texture edge segment and all the edge curvatures corresponding to the texture edge segment; Continue to determine the edge labels of the remaining texture edge segments; Divide the bottom sediment area without aquatic organisms and multiple potential biological areas to be verified in the underwater image according to the edge labels of all texture edge segments.

7. The method according to claim 1, characterized in that, Determining the color texture features of the texture structure of each potential biological area based on the color space distance and the gray-level co-occurrence matrix of each potential biological area specifically includes: Obtain the standardized image of the underwater image, and convert the standardized image into a grayscale processed image; Intercept the grayscale sub-image corresponding to each potential biological area in the grayscale processed image; Determine the gray-level co-occurrence matrix of each grayscale sub-image; Extract the texture feature parameters of each potential biological area from the gray-level co-occurrence matrix; Construct the color texture features of the texture structure of each potential biological area based on the texture feature parameters and the color space distance.

8. The method according to claim 1, characterized in that, Further extracting multiple biological candidate areas from all potential biological areas according to the color texture features and the contour features of each potential biological area specifically includes: Determine the contour features of each potential biological area; Call the pre-trained classifier model; ​ 9. The method according to claim 1, wherein ​ Select a biological candidate area as the selected biological candidate area, and obtain the gradient feature sequence of each pixel point in the selected biological candidate area; Extract a plurality of texture breakpoints from all pixel points according to the gradient feature sequence of each pixel point; Determine the spatial distribution entropy of all texture breakpoints in the selected biological candidate area; Continue to determine the spatial distribution entropy of all texture breakpoints in the remaining biological candidate areas.

10. An aquatic biodiversity monitoring system based on image recognition, which uses the method described in any one of claims 1 to 9 for image recognition. The system includes an image recognition unit, and is characterized in that, The image recognition unit includes: An acquisition module, configured to acquire an underwater image of a monitored water area, and determine a bottom substrate area without aquatic organisms and a plurality of potential biological areas to be verified in the underwater image; A processing module, configured to respectively determine the color space distance between each potential biological area and the bottom substrate area, and determine the color texture feature of the texture structure of each potential biological area based on the color space distance in combination with the gray level co-occurrence matrix of each potential biological area; The processing module is further configured to further extract a plurality of biological candidate areas from all potential biological areas according to the color texture feature and the contour feature of each potential biological area, and determine the spatial distribution entropy of all texture breakpoints in each biological candidate area; An execution module, configured to determine the mimicry label of the mimicry aquatic organisms in each biological candidate area based on the spatial distribution entropy and texture structure of all texture breakpoints in each biological candidate area, and generate a dynamic map for monitoring aquatic biodiversity according to all the mimicry labels.

Images