Method for identifying the use of interception nets for deep-sea underwater robots

Through night vision infrared camera acquisition and deep learning model recognition, combined with image preprocessing and feature extraction, the problem of deep-sea interceptor network detection is solved, and the accurate identification and monitoring of the use of interceptor network is achieved, which is suitable for deep sea environments.

CN118762236BActive Publication Date: 2024-12-06TIANJIN PORT ENG INST LTD OF CCCC FIRST HARBOR ENG +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411224134.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-12-06
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

Due to the light-free environment and energy supply difficulties of deep and far-sea submersible interceptor networks, it is difficult for the existing technology to effectively detect dirt blockage, corrosion or damage.

Method used

The night vision infrared camera is used to collect interceptor network images, and it is recognized through image preprocessing, feature extraction and deep learning model, including texture, shape and roughness feature extraction, combining convolutional layer, pooling layer, feature pyramid network and full connection layer. Identify the usage of the interceptor network.

Benefits of technology

It realizes the accurate identification of the interceptor network in a deep-sea environment, dirt blockage, corrosion or damage, and provides efficient monitoring and maintenance solutions, suitable for deep-sea conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118762236B_ABST
    Figure CN118762236B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for identifying the use of an interception net of a deep-sea underwater robot. The method uses a camera with a night vision infrared function to obtain a clear image of an interception net in a deep-sea lightless environment. Then, a training data set is constructed through image preprocessing, image segmentation, and image feature extraction to train a deep-sea interception net use recognition deep learning model. The deep learning model for identifying the use of a deep-sea interception net first performs convolution processing on the training set through its convolution layer to extract a feature map, then performs pooling layer processing to obtain a feature map with reduced resolution, then uses a feature pyramid network to obtain a fused multi-scale feature map, then performs deformable convolution layer processing to obtain a feature map after learning displacement, then performs pooling layers of different sizes to obtain a fused multi-scale pooling feature map, then outputs a classification result through a fully connected layer, and finally outputs: clean and intact, blocked by dirt, corroded or damaged, etc., the use of the interception net.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of underwater robots and image recognition, and in particular relates to a method for identifying the use of an interception net of a deep-sea underwater robot. Background Art

[0002] Underwater interception nets are a type of underwater protection and interception device. By placing interception nets in specific underwater areas, the physical blocking effect of the nets is used to prevent the target objects from passing through. When an object touches the interception net, the corresponding monitoring system will be triggered so that timely response measures can be taken. For example, setting up underwater interception nets around facilities such as offshore oil platforms can prevent large marine organisms or foreign objects from causing damage to the facilities.

[0003] Underwater interception nets are usually made of high-strength synthetic fiber materials. The shape and size of the net can be designed and customized according to specific usage requirements. Fixtures are usually installed to ensure its stability underwater. During long-term use underwater, underwater interception nets are faced with the possibility of being blocked by dirt, corroded or damaged. How to conveniently, effectively and accurately detect whether underwater interception is blocked by dirt, corroded or damaged is a technical problem that needs to be solved. Especially for underwater interception nets in deep and far sea working scenarios, traditional visual detection methods are difficult to work because the light in the deep sea is almost zero; and in deep and far environments, traditional energy supply methods are difficult to achieve. Summary of the invention

[0004] The present invention mainly aims to overcome the deficiencies of the prior art and provides a method for identifying the use of a deep-sea underwater robot interception net.

[0005] The present invention is achieved through the following technical solutions:

[0006] A method for identifying the use of a deep-sea underwater robot interception net comprises the following steps:

[0007] Step 1: Collect infrared images of the underwater interception net;

[0008] Step 2, preprocessing the image collected in step 1;

[0009] Step 3, performing image segmentation on the image preprocessed in step 2;

[0010] Step 4, extracting features of the image processed in step 3;

[0011] Image feature extraction, including: texture feature extraction, shape feature extraction and roughness feature extraction;

[0012] Texture feature extraction: Use the local binary pattern method to describe the texture features of the image by comparing the pixel value with the pixel values ​​of its surrounding neighborhood to identify corrosion and dirt;

[0013] The formula for texture feature extraction is:

[0014] ;

[0015] in, Indicates at location LBP value; is the number of neighborhood pixels; is the radius of the neighborhood; is the neighborhood pixel value; is the center pixel value; is a symbolic function, defined as follows:

[0016] ;

[0017] Shape feature extraction: Use edge detection to extract the shape features of the object, and extract the edge of the object by detecting the gradient change of the image, which is used to identify whether the mesh of the underwater interception net is blocked by dirt;

[0018] The shape feature extraction process is as follows:

[0019] (1) Perform Gaussian filtering: ;

[0020] in, is the image processed by step 2; is the Gaussian kernel; is the image after Gaussian filtering;

[0021] (2) Calculate the gradient of the image:

[0022] ;

[0023] in, are the horizontal and vertical gradients of the image, respectively; x Represents the horizontal direction of the image, y Represents the vertical direction of the image;

[0024] (3) Calculate the gradient magnitude and direction:

[0025] ;

[0026] ;

[0027] in, is the gradient amplitude; is the gradient direction;

[0028] (4) Perform non-maximum suppression:

[0029] By comparing the pixel point with its neighboring pixel values ​​in the gradient direction, the local maximum point is retained;

[0030] (5) Compare the gradient magnitude with the set high and low thresholds to determine the edge in the image:

[0031] ;

[0032] in, Indicates the high threshold, Indicates a low threshold; when the gradient amplitude When the gradient amplitude is greater than the high threshold, it is marked as a strong edge point, represented by 1; When the gradient amplitude is less than the low threshold, it is marked as a non-edge point and represented by 0; When it is between the high threshold and the low value, it is marked as a weak edge point, represented by weak edge, and whether to keep it is determined by connecting the strong edge points;

[0033] Roughness feature extraction: Calculate the surface roughness of the image to identify corrosion areas;

[0034] Surface roughness of the image The calculation formula is: ;

[0035] in, is the number of pixels; is the gray value of each pixel; is the average gray value;

[0036] Step 5: According to the above steps 1 to 4, the collected infrared images of interception nets, including clean and intact, blocked by dirt, corroded, and damaged, are processed to obtain the feature matrix of the image including texture features, shape features, and roughness features. These feature matrices and the images processed in step 3 are integrated into a matrix set as a training set to train a deep learning model for identifying the use of deep-sea interception nets;

[0037] The network structure of the deep sea interception network usage identification deep learning model includes:

[0038] (1) Convolutional layer: Its input is the training set obtained in step 5, and its output is the feature map extracted by convolution;

[0039] (2) Pooling layer: Its input is the feature map obtained by the convolution layer, and its output is a feature map with reduced resolution;

[0040] (3) Feature pyramid network: Its input is the feature map obtained by pooling layer processing, and its output is the fused multi-scale feature map;

[0041] (4) Deformable convolution layer: Its input is the multi-scale feature map obtained by feature pyramid network processing, and its output is the feature map after learning displacement;

[0042] (5) Pooling layers of different sizes: The input is the feature map obtained by the deformable convolution layer, and the output is the fused multi-scale pooling feature map;

[0043] (6) Fully connected layer: Its input is the feature map obtained by processing the pooling layers of different sizes, and its output is the classification result;

[0044] (7) Output layer: Its input is the classification result obtained through the fully connected layer, and its output is the final recognition result, which includes: clean, blocked by dirt, corroded, or damaged.

[0045] In the above technical solution, in step one, an underwater robot is used, which is equipped with a night vision infrared camera. The underwater robot moves along a preset path to collect images of the interception net in the deep sea lightless environment, stores the captured images in real time, and transmits them to the processing module wirelessly or wiredly; photovoltaic panels are arranged on the water surface or in a shallow underwater layer, and photovoltaic power generation is used to provide power for the underwater robot.

[0046] In the above technical solution, in step 2, image preprocessing includes: bilateral filtering processing and adaptive histogram equalization processing.

[0047] In the above technical solution, the bilateral filtering process is used for smoothing the image and retaining edge details;

[0048] The formula for bilateral filtering is:

[0049] ;

[0050] in, is the original image; is the image after bilateral filtering; p and q is the position vector of the pixel in the image, To represent the position of a specific pixel in the image, is the row index of the pixel, j is the column index of the pixel; is a spatial domain parameter that controls the influence of spatial distance, generally between; is the gray value range parameter, which controls the influence of gray value difference, generally between; is the normalization coefficient, ensuring that the sum of weights is 1; What is calculated is the spatial distance between two pixels in the image, that is, how far apart the two pixels are in the image; Refers to pixels p and pixels q The square of the difference in pixel values, and Respectively represent the pixel positions p and q The pixel value at .

[0051] In the above technical solution, the adaptive histogram equalization processing is used to enhance image contrast and improve detail clarity;

[0052] The formula for adaptive histogram equalization is:

[0053] ;

[0054] in, is the image after bilateral filtering; is the image after adaptive histogram equalization; is the number of gray levels, usually 256; and exist between; is the minimum pixel value of all pixels (if the pixel range of the image is ,So is 0, which means that at least one pixel in the image has a grayscale value of 0); is the maximum pixel value of all pixels (if the pixel range of the image is ,So is 255, which means that at least one pixel in the image has a grayscale value of 255.

[0055] In the above technical solution, in step 3, the Otsu threshold is used to automatically select the optimal threshold for image segmentation;

[0056] The image segmentation formula is:

[0057] ;

[0058] in, is the ratio of foreground and background pixels; is the mean grayscale value of foreground and background pixels; is the total grayscale mean of the image.

[0059] The advantages and beneficial effects of the present invention are:

[0060] The present invention uses a camera with night vision infrared function to obtain clear images of interception nets in deep-sea lightless environments, and then constructs a training data set through image preprocessing, image segmentation and image feature extraction to train a deep learning model for identifying the use of deep-sea interception nets. The deep learning model for identifying the use of deep-sea interception nets first extracts feature maps by convolution processing on the training set through its convolution layer, then obtains feature maps with reduced resolution through pooling layer processing, and then uses feature pyramid network processing to obtain fused multi-scale feature maps, and then obtains feature maps after learning displacement through deformable convolution layer processing, and then obtains fused multi-scale pooling feature maps through pooling layers of different sizes, and then outputs classification results through the fully connected layer, and finally outputs the final recognition results, including: clean and intact, blocked by dirt, corroded or damaged interception net usage.

[0061] The method of the present invention can use an underwater robot equipped with a night vision infrared camera, and the underwater robot moves along a preset path to collect images of the interception net in a deep-sea lightless environment, store the captured images in real time, and transmit them to a processing module by wireless or wired means; in addition, photovoltaic panels are arranged on the water surface or in a shallow underwater layer, and photovoltaic power generation is used to provide power for the underwater robot. This implementation method is highly feasible and practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of the method for identifying the use of the deep-sea underwater robot interception net of the present invention.

[0063] Figure 2 This is the network structure diagram of the deep learning model for identifying the usage of deep-sea interception networks.

[0064] For ordinary technicians in this field, other relevant drawings can be obtained based on the above drawings without any creative work. DETAILED DESCRIPTION

[0065] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with specific embodiments.

[0066] The present invention designs a method for identifying the use of deep-sea underwater robot interception nets, see the attached Figure 1 , including the following steps.

[0067] Step 1: Collect infrared images of the underwater interception net.

[0068] Specifically, an underwater robot can be used, equipped with a night vision infrared camera, and the underwater robot moves along a preset path to collect images of the interception net in the deep sea lightless environment, store the captured images in real time, and transmit them to the processing module wirelessly or wiredly. Photovoltaic panels can also be arranged on the water surface or in shallow water to use photovoltaic power generation to provide power for the underwater robot.

[0069] Step 2: preprocess the image collected in step 1.

[0070] Image preprocessing, including: bilateral filtering and adaptive histogram equalization.

[0071] (1) Bilateral filtering: used to smooth images and retain edge details.

[0072] The formula for bilateral filtering is:

[0073] ;

[0074] in, is the original image (the image is represented by a two-dimensional matrix, and each element of the matrix corresponds to a pixel in the image); is the image after bilateral filtering; p and q is the position vector of the pixel in the image, To represent the position of a specific pixel in the image, is the row index of the pixel, j is the column index of the pixel; is a spatial domain parameter that controls the influence of spatial distance, generally between; is the gray value range parameter, which controls the influence of gray value difference, generally between; is the normalization coefficient, ensuring that the sum of weights is 1; What is calculated is the spatial distance between two pixels in the image, that is, how far apart the two pixels are in the image; Refers to pixels p and pixels q The square of the difference in pixel values, and Respectively represent the pixel positions p and q The pixel value at .

[0075] (2) Adaptive histogram equalization processing: used to enhance image contrast and improve detail clarity.

[0076] The formula for adaptive histogram equalization is:

[0077] ;

[0078] in, is the image after bilateral filtering; is the image after adaptive histogram equalization; is the number of gray levels, usually 256; and exist between; is the minimum pixel value of all pixels (if the pixel range of the image is ,So is 0, which means that at least one pixel in the image has a grayscale value of 0); is the maximum pixel value of all pixels (if the pixel range of the image is ,So is 255, which means that at least one pixel in the image has a grayscale value of 255.

[0079] Step three, the image preprocessed in step two is segmented to reduce the computational complexity of subsequent processing steps by focusing on specific areas or objects in the image, and the Otsu threshold is used to automatically select the optimal threshold for image segmentation.

[0080] The image segmentation formula is:

[0081] ;

[0082] in, is the ratio of foreground and background pixels; is the mean grayscale value of foreground and background pixels; is the total grayscale mean of the image.

[0083] Relationship between parameters and recognition results: Otsu threshold can automatically select the best segmentation point and improve the accuracy of image segmentation.

[0084] Step 4: extract image features from the image processed in step 3.

[0085] Image feature extraction, including: texture feature extraction, shape feature extraction and roughness feature extraction.

[0086] (1) Texture feature extraction: The Local Binary Patterns (LBP) method is used to describe the texture features of the image by comparing the pixel value with the pixel values ​​of its surrounding neighborhood to identify corrosion and dirt.

[0087] The formula for texture feature extraction is:

[0088] ;

[0089] in, Indicates at location LBP value; is the number of neighborhood pixels, common values ​​are 8, 16, etc.; is the radius of the neighborhood, common values ​​are 1, 2, 3, etc.; is the neighborhood pixel value; is the center pixel value; is a symbolic function, defined as follows:

[0090] ;

[0091] The relationship between the parameters and the recognition results is:

[0092] Larger and More texture information can be captured, but the computational complexity will also increase.

[0093] The histogram of LBP values ​​is used to describe the texture features of the image. By analyzing the histogram, it is possible to identify whether there is corrosion and dirt in the image.

[0094] (2) Shape feature extraction: The shape features of the object are extracted using edge detection. The edges of the object are extracted by detecting the gradient changes in the image to identify whether the mesh of the underwater interception net is blocked by dirt.

[0095] The shape feature extraction process is as follows:

[0096] 1. Perform Gaussian filtering: ;

[0097] in, is the image processed by step 2; is the Gaussian kernel; is the image after Gaussian filtering (smoothing / denoising);

[0098] 2. Calculate the gradient of the image:

[0099] ;

[0100] in, are the horizontal and vertical gradients of the image, respectively; x Represents the horizontal direction of the image, y Represents the vertical direction of the image;

[0101] 3. Calculate the gradient magnitude and direction:

[0102] ;

[0103] ;

[0104] in, is the gradient amplitude; is the gradient direction;

[0105] 4. Perform non-maximum suppression:

[0106] By comparing the pixel point with its neighboring pixel values ​​in the gradient direction, the local maximum point is retained;

[0107] 5. Compare the gradient magnitude with the set double thresholds (high threshold and low threshold) to determine the edge in the image:

[0108] ;

[0109] in, Indicates the high threshold, Indicates a low threshold; when the gradient amplitude When the gradient amplitude is greater than the high threshold, it is marked as a strong edge point, represented by 1; When the gradient amplitude is less than the low threshold, it is marked as a non-edge point and represented by 0; When it is between the high threshold and the low value, it is marked as a weak edge point, represented by weakedge, and then it is decided whether to keep it by connecting the strong edge points;

[0110] The high threshold is generally taken at the high percentile of the gradient amplitude, such as ; The low threshold is usually taken at the high threshold to .

[0111] Relationship between parameters and recognition results:

[0112] The choice of dual thresholds directly affects the accuracy and robustness of edge detection. If the high threshold is set too high, some real edges may not be detected, while if the low threshold is set too low, more noise edges will be introduced.

[0113] The appropriate dual threshold combination can effectively separate edges from non-edges, ensuring the accuracy and robustness of edge detection.

[0114] (3) Roughness feature extraction: Calculate the surface roughness of the image to identify corrosion areas.

[0115] Surface roughness of the image The calculation formula is:

[0116] ;

[0117] in, is the number of pixels; is the gray value of each pixel; is the average gray value;

[0118] Step five, according to the above steps one to four, the collected infrared images of interception nets, including clean and intact, blocked by dirt, corroded, and damaged, are processed to obtain the feature matrix of the image containing texture features, shape features, and roughness features. These feature matrices and the images processed in step three are integrated into a matrix set as a training set, and the deep learning model for identifying the usage of deep-sea interception nets is trained.

[0119] The network structure of the deep learning model for identifying the usage of the deep sea interception network is as follows: Figure 2 As shown, the input, output and functions of each layer of the network are as follows:

[0120] (1) Convolutional layer: Its input is the training set obtained in step 5, and its output is the feature map extracted by convolution (which is different from the concept of texture features, shape features, and roughness features in the input. The feature map here is the weight and bias of the neural network, which will not be repeated hereafter); its function is to extract low-level and high-level features;

[0121] (2) Pooling layer: Its input is the feature map obtained by convolution layer processing, and its output is the feature map with reduced resolution. Its function is to reduce the size of feature map and extract important features.

[0122] (3) Feature pyramid network: Its input is the feature map obtained by pooling layer processing, and its output is the fused multi-scale feature map; its function is to fuse multi-scale features to improve detection accuracy;

[0123] (4) Deformable convolution layer: Its input is a multi-scale feature map obtained by processing the feature pyramid network, and its output is a feature map after learning displacement; its function is to adapt to geometric changes and improve recognition ability;

[0124] (5) Pooling layers of different sizes: The input is the feature map obtained by the deformable convolution layer, and the output is the fused multi-scale pooling feature map; the function is to provide multi-scale context information and improve the feature representation capability;

[0125] (6) Fully connected layer: Its input is the feature map obtained by processing the pooling layers of different sizes, and its output is the classification result; its function is to perform classification or regression operations;

[0126] (7) Output layer: Its input is the classification result obtained through the fully connected layer, and its output is the final recognition result. Its function is to output the recognition result, which includes: clean and intact, clogged with dirt, corroded or damaged.

[0127] The present invention is described above by way of example. It should be noted that, without departing from the core of the present invention, any simple deformation, modification or other equivalent replacement that can be made by those skilled in the art without inventive effort falls within the protection scope of the present invention.

Claims

1. A method for identifying the use of deep-sea underwater robot interception nets, It is characterized in that The following steps are involved: Step 1: Collect infrared images of the underwater interception net; Step 2, preprocessing the image collected in step 1; Step 3, performing image segmentation on the image preprocessed in step 2; Step 4, extracting features of the image processed in step 3; Image feature extraction, including: texture feature extraction, shape feature extraction and roughness feature extraction; Texture feature extraction: Use the local binary pattern method to describe the texture features of the image by comparing the pixel value with the pixel values ​​of its surrounding neighborhood to identify corrosion and dirt; The formula for texture feature extraction is: Among them, LBP P,R (x, y) represents the LBP value at position (x, y); P is the number of neighborhood pixels; R is the radius of the neighborhood; i p is the neighborhood pixel value; i c is the center pixel value; s(i p -i c ) is a symbolic function, defined as follows: Shape feature extraction: Use edge detection to extract the shape features of the object, and extract the edge of the object by detecting the gradient change of the image, which is used to identify whether the mesh of the underwater interception net is blocked by dirt; The shape feature extraction process is as follows: (1), perform Gaussian filtering: I′=I*G; Where I is the image processed in step 2; G is the Gaussian kernel; I ′ is the image after Gaussian filtering; (2) Calculate the gradient of the image: Among them, G x , G y They are the horizontal and vertical gradients of the image respectively; x represents the horizontal direction of the image, and y represents the vertical direction of the image; (3), calculate the gradient magnitude and direction: Where G is the gradient amplitude; θ is the gradient direction; (4) Perform non-maximum suppression: By comparing the pixel point with its neighboring pixel values ​​in the gradient direction, the local maximum point is retained; (5) Compare the gradient magnitude with the set high and low thresholds to determine the edge in the image: Among them, high threshold represents the high threshold, and low threshold represents the low threshold; when the gradient amplitude G is greater than the high threshold, it is marked as a strong edge point, represented by 1; when the gradient amplitude G is less than the low threshold, it is marked as a non-edge point, represented by 0; when the gradient amplitude G is between the high threshold and the low threshold, it is marked as a weak edge point, represented by weak edge, and the subsequent connection is used to determine whether to retain it; Roughness feature extraction: Calculate the surface roughness of the image to identify corrosion areas; The surface roughness R of the image q The calculation formula is: Where N is the number of pixels; z i is the gray value of each pixel; is the average gray value; Step 5: According to the above steps 1 to 4, the collected infrared images of interception nets, including clean and intact, blocked by dirt, corroded, and damaged, are processed to obtain feature matrices of the images including texture features, shape features, and roughness features. These feature matrices and the images processed in step 3 are integrated into a matrix set as a training set to train a deep learning model for identifying the use of deep-sea interception nets; The network structure of the deep sea interception network usage identification deep learning model includes: (1) Convolutional layer: Its input is the training set obtained in step 5, and its output is the feature map extracted by convolution; (2) Pooling layer: Its input is the feature map obtained by the convolution layer, and its output is a feature map with reduced resolution; (3) Feature pyramid network: Its input is the feature map obtained by pooling layer processing, and its output is the fused multi-scale feature map; (4) Deformable convolutional layer: Its input is the multi-scale feature map obtained by feature pyramid network processing, and its output is the feature map after learning displacement; (5) Pooling layers of different sizes: Its input is the feature map obtained by the deformable convolution layer, and its output is the fused multi-scale pooling feature map; (6) Fully connected layer: Its input is the feature map obtained by pooling layers of different sizes, and its output is the classification result; (7) Output layer: Its input is the classification result obtained by the fully connected layer, and its output is the final recognition result, which includes: clean, blocked by dirt, corroded or damaged.

2. The method for identifying the use of a deep-sea underwater robot interception net according to claim 1, Features: In step one, an underwater robot is used, which is equipped with a night vision infrared camera. The underwater robot moves along a preset path, collects images of the interception net in the deep sea lightless environment, stores the captured images in real time, and transmits them to the processing module wirelessly or wiredly; photovoltaic panels are arranged on the water surface or in the shallow underwater layer, and photovoltaic power generation is used to provide power for the underwater robot.

3. The method for identifying the use of a deep-sea underwater robot interception net according to claim 1, Features: In step 2, image preprocessing includes bilateral filtering and adaptive histogram equalization.

4. The method for identifying the use of a deep-sea underwater robot interception net according to claim 3, Features: The bilateral filtering process is used for smoothing the image and retaining edge details. The formula for bilateral filtering is: Where I1 is the original image; I2 is the image after bilateral filtering; p and q are the position vectors of the pixels in the image, (i, j) represents the position of a specific pixel in the image, i is the row index of the pixel, and j is the column index of the pixel; σ s is a spatial domain parameter that controls the influence of spatial distance and is between [1,10]; σ r is the grayscale range parameter, which controls the influence of grayscale value difference, between [0.1, 0.5]; Z(x, y) is the normalization coefficient, which ensures that the weight sum is 1; ∥pq∥ 2 What is calculated is the spatial distance between two pixels in the image, that is, how far apart the two pixels are in the image; ∥I1(p)-I1(q)∥ 2 refers to the square of the difference in pixel values ​​between pixel p and pixel q, I1(p) and I1(q) represent the pixel values ​​at pixel positions p and q, respectively.

5. The method for identifying the use of the deep-sea underwater robot interception net according to claim 3, Features: The adaptive histogram equalization process is used to enhance image contrast and improve detail clarity; The formula for adaptive histogram equalization is: Among them, I2 is the image after bilateral filtering; I3 is the image after adaptive histogram equalization; L is the number of gray levels; I2 and I3 are between [0,255]; min(I) is the minimum pixel value of all pixels; max(I) is the maximum pixel value of all pixels.

6. The method for identifying the use of a deep-sea underwater robot interception net according to claim 1, Features: In step 3, the Otsu threshold is used to automatically select the optimal threshold for image segmentation; The image segmentation formula is: Among them, ω 0 ,ω 1 is the ratio of foreground and background pixels; μ 0 , μ 1 is the mean grayscale value of foreground and background pixels; μ T is the total grayscale mean of the image.