Underwater feature recognition and tracking research based on deep learning

Abstract

The invention relates to the technical field of underwater image processing, in particular to underwater feature recognition and tracking research based on deep learning, and aims to improve the image processing quality through analysis from the perspective of image information processing. Meanwhile, an algorithm combining adaptive contrast enhancement and contrast-limited adaptive histogram equalization is adopted to improve the image contrast and highlight feature edge information, and denoising processing is performed on the feature-enhanced image through a denoising template method, so that noise interference in the underwater image is effectively suppressed, and the image quality is improved. The method is advantaged in that data support is provided for subsequent feature identification, improvement of underwater feature identification precision is facilitated, underwater welding seam feature information is extracted from the filtering optimization feature image based on the optimal YOLOv8 network model, feature identification precision and accuracy are further improved, laser sensor attitude calibration is carried out through a feature information mapping mode, and accuracy of underwater welding seam feature identification is further improved. And the feature tracking accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater image processing, and in particular to a study on underwater feature recognition and tracking based on deep learning. Background Art

[0002] Welding defects are the most difficult form of surface damage to detect. They have characteristics such as cracks, pores, slag inclusions, etc., and are extremely difficult to detect and identify. When conducting underwater target detection, especially for underwater hull detection, considering the effect of water, relying solely on information collected by a single sensor cannot obtain accurate welding defect detection results.

[0003] Among them, underwater optical images are often affected by noise interference, blurred texture features, low contrast and color distortion, which seriously affect the effect of image recognition. However, in the prior art, it is impossible to effectively process underwater optical images, which leads to poor image processing effect, is not conducive to the extraction and recognition of underwater features, and thus reduces the extraction and recognition efficiency of underwater features, and at the same time affects the subsequent feature tracking trajectory generation accuracy, reduces the automation efficiency of underwater machines and the safety of feature tracking trajectories;

[0004] In view of the above technical defects, a solution is now proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a research on underwater feature recognition and tracking based on deep learning to solve the above-mentioned technical defects. The present invention analyzes from the perspective of image information processing to improve the image processing quality, and extracts underwater weld feature information from the filtered optimized feature image based on the YOLOv8 network model, thereby improving the feature recognition precision and accuracy, and calibrates the laser sensor posture by feature information mapping to provide technical support for subsequent feature tracking to improve feature tracking accuracy, and performs actual tracking self-inspection on the generated feature tracking trajectory by information feedback, so as to timely adjust the real-time moving trajectory of the welding robot to improve the movement accuracy of the welding robot, and further manage and adjust the feature tracking trajectory generation scheme to reduce the risk of obstacle avoidance failure of the welding robot.

[0006] The purpose of the present invention can be achieved by the following technical solution: A study on underwater feature recognition and tracking based on deep learning, comprising the following steps:

[0007] S1, scanning underwater features, capturing underwater feature contour information, and setting the underwater feature contour information as a feature image to be processed;

[0008] S2, the industrial computer performs filtering: the collected feature image to be processed is subjected to image enhancement processing by using an algorithm combining adaptive contrast enhancement and limited contrast adaptive histogram equalization (CLAHE) to obtain an enhanced feature image;

[0009] S3, denoising: denoising the enhanced feature image using a denoising template to obtain a filtered optimized feature image;

[0010] S4: Build a deep learning model and set model training parameters, run the YOLOv8 network model to extract underwater weld feature information from the filtered optimized feature image;

[0011] S5: Accurately map the underwater weld feature information to the coordinate system of the underwater machine to calibrate the laser sensor posture;

[0012] S6: After posture calibration, the RRT (rapid random tree) path planning algorithm is used to generate a feature tracking trajectory. The controller drives the welding robot to guide the movement of the welding gun according to the feature tracking trajectory, performs deviation warning feedback analysis on the feature tracking trajectory of the welding robot, performs discrimination processing on the obtained avoidance deviation index, and outputs the obtained adjustment instruction as feedback.

[0013] Preferably, the process of acquiring the enhanced feature image is as follows:

[0014] The feature image to be processed is divided into a high-frequency image and a low-frequency image through a template filtering operation, and the high-frequency image is further enhanced. The enhanced high-frequency image and the low-frequency image are fused through an image enhancement algorithm to obtain an enhanced feature image. The mathematical expression of the image enhancement algorithm is shown in the following formula: g(x, y) = m d (x,y)+G(x,y)[f(x,y)-m d (x,y)], where g(x,y) represents the grayscale value of the processed image at point (x,y), and f(x,y) represents the grayscale value of the image before processing at point (x,y); m d (x, y) is the low-frequency image part obtained after filtering, G(x, y) is the dynamic gain function, and the G(x, y) value can change with the change of the processed pixel points;

[0015] The mean filtering method is used as a means to obtain a low-frequency image. The specific mathematical expression is as follows: The ratio of the image grayscale mean and the standard deviation in the neighborhood is used as the dynamic gain, and the following formula is established: Where σ(x,y) is defined as follows: The template radius n is artificially set to 2.

[0016] Preferably, the denoising template is obtained in the following manner:

[0017] The denoising template image is obtained by the following formula: Among them, M(x,y) is the image element value of the denoising template, C(x,y) is the grayscale value of the image pixel after the closing operation, k is the noise judgment threshold, and k is a constant. The purpose of this operation is to compare the difference in pixel grayscale between the enhanced feature image and the enhanced feature image denoised by the closing operation, and mark the granular black spots in the enhanced feature image whose pixel grayscale difference between the enhanced feature image and the enhanced feature image denoised by the closing operation is greater than the preset threshold as potential noise points; the grayscale value of the potential noise point is replaced by the grayscale information of other points through the median filtering method, and the original enhanced feature image is updated after each noise filtering operation on a potential noise point, and finally the filtered optimized feature image is obtained.

[0018] Preferably, the YOLOv8 network model construction process is as follows:

[0019] Setting model training parameters, building a data set based on the model training parameters, and dividing the data set into a training set and a validation set. The training set is used to teach the neural network model, while the validation set is responsible for verifying the effectiveness of the model learning results.

[0020] Run the model and start monitoring the entire training process. The main things that need to be modified are the settings of the training rounds, batch size, and learning rate. Monitor the changes in the training set bounding box loss, target loss, and classification loss during the training process, and ultimately help screen out the YOLOv8 network model with the best performance.

[0021] Preferably, the underwater weld feature information is accurately mapped to the coordinate system of the underwater machine so as to calibrate the laser sensor posture. The specific process is as follows:

[0022] Define two coordinate systems: {S} represents the sensor coordinate system, {B} represents the reference coordinate system, and the coordinates of point P in these two coordinate systems are expressed as B P and S P;

[0023] Determine the transformation matrix of the sensor coordinate system {S} relative to the end coordinate system {E} For any known point in its coordinate system S Q, the formula to transform it into the machine base coordinate system is:

[0024] Preferably, the transformation matrix of the sensor coordinate system {S} relative to the end coordinate system {E} is Calculate the transformation matrix The specific steps are as follows:

[0025] SS1: Determine a specific point P on the target and record the position of point P in the reference coordinate system {B} B P=(x B ,y B ,z B ,1) T ;

[0026] SS2: Adjust the position of the machine to ensure that the laser line emitted by the sensor can pass through point P. Also record the position of point P in the sensor coordinate system {S} S P=(x S ,0,z S ,1) T ;

[0027] SS3: Switch the current tool coordinate system of the machine to the end coordinate system {E}, record the pose data in the end coordinate system {E} at this time, and use Euler's rotation theorem to calculate the rotation matrix Further seek SS4: According to the transformation relationship of point P in space:

[0028] For any known point in its coordinate system S Q, the formula to transform it into the machine base coordinate system is: In the formula, B Q is the position coordinate of point Q in the reference coordinate system {B}, S Q is the position coordinate of point Q in the sensor coordinate system {S}.

[0029] Preferably, the deviation warning feedback analysis process is as follows:

[0030] The real-time moving trajectory of the welding robot is obtained, and the real-time moving trajectory is compared with the generated feature tracking trajectory for coincidence, thereby obtaining the coincidence between the real-time moving trajectory and the generated feature tracking trajectory, and setting the coincidence between the real-time moving trajectory and the generated feature tracking trajectory as the moving deviation warning coefficient, and performing discrimination processing on the moving deviation warning coefficient to generate a feedback instruction or generate an alarm instruction;

[0031] When generating a feedback instruction, the generated feature tracking trajectory is divided into an avoidance tracking trajectory and a non-avoidance tracking trajectory, the real-time avoidance trajectory of the welding robot is obtained, the minimum avoidance distance of the avoidance tracking trajectory of the welding robot is obtained, the minimum avoidance distance represents the minimum straight-line distance between the obstacle and the welding robot, and the actual minimum avoidance distance of the real-time avoidance trajectory of the welding robot is obtained. The value obtained by subtracting the minimum avoidance distance from the actual minimum avoidance distance is set as the avoidance deviation index, and the avoidance deviation index is judged: if the avoidance deviation index is greater than or equal to the preset avoidance deviation index threshold, no signal is generated; if the avoidance deviation index is less than the preset avoidance deviation index threshold, an adjustment instruction is generated.

[0032] The beneficial effects of the present invention are as follows:

[0033] (1) The present invention analyzes from the perspective of image information processing to improve the image processing quality, and adopts an algorithm combining adaptive contrast enhancement and limited contrast adaptive histogram equalization to enhance image contrast and highlight feature edge information, and denoises the enhanced feature image by a denoising template method, thereby effectively suppressing noise interference in underwater images, so as to provide data support for subsequent feature recognition;

[0034] (2) The present invention extracts underwater weld feature information from the filtered optimized feature image based on the YOLOv8 network model, thereby improving the feature recognition precision and accuracy, and calibrates the laser sensor posture by means of feature information mapping to provide technical support for subsequent feature tracking, so as to improve the feature tracking accuracy, and performs actual tracking self-inspection on the generated feature tracking trajectory by means of information feedback, so as to timely adjust the real-time moving trajectory of the welding robot to improve the movement accuracy of the welding robot, and further manage and adjust the feature tracking trajectory generation scheme to reduce the risk of obstacle avoidance failure of the welding robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below in conjunction with the accompanying drawings;

[0036] Figure 1 is a reference diagram of the method of the present invention;

[0037] Figure 2 It is a local analysis reference diagram of the present invention. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Embodiment 1:

[0040] See also Figure 1 to Figure 2 As shown, the present invention is a research on underwater feature recognition and tracking based on deep learning, comprising the following steps:

[0041] S1, scanning underwater features, capturing underwater feature contour information, and setting the underwater feature contour information as a feature image to be processed;

[0042] S2, the industrial computer performs filtering: the collected feature image to be processed is subjected to image enhancement processing by using an algorithm combining adaptive contrast enhancement and limited contrast adaptive histogram equalization (CLAHE) to obtain an enhanced feature image;

[0043] S3, denoising: denoising the enhanced feature image using a denoising template to obtain a filtered optimized feature image;

[0044] S4: Build a deep learning model and set model training parameters, run the YOLOv8 network model to extract underwater weld feature information from the filtered optimized feature image;

[0045] S5: Accurately map the underwater weld feature information to the coordinate system of the underwater machine to calibrate the laser sensor posture;

[0046] S6: After posture calibration, the RRT (rapid random tree) path planning algorithm is used to generate a feature tracking trajectory. The controller drives the welding robot to guide the movement of the welding gun according to the feature tracking trajectory, performs deviation warning feedback analysis on the feature tracking trajectory of the welding robot, performs discrimination processing on the obtained avoidance deviation index, and outputs the obtained adjustment instruction as feedback.

[0047] In the embodiment of the present invention: S1: scanning underwater features, capturing underwater feature contour information, and setting the underwater feature contour information as a feature image to be processed;

[0048] Specifically, the system uses a laser sensor installed at the end of the robot arm to scan underwater features. That is, the laser sensor scans the weldment to collect the contour information of the weld groove. The main goal of the scan is to capture the initial data of the edge of the underwater feature, which is the prerequisite for accurate underwater feature tracking.

[0049] The laser sensor uses triangulation technology to accurately extract the distance information of the target object and outlines the contour details of the object through detailed analysis of the scan data;

[0050] When performing scanning operations, the laser sensor is firmly mounted at the end of the underwater robot arm and maintained in a parallel posture with the underwater robot arm to ensure that the projected laser line can accurately hit the detection surface at a 90-degree angle. This design is intended to fully cover and record the underwater feature morphology in detail; by continuously manipulating the movement of the machine to continuously collect data, and then splicing it into a complete morphological map of the tracking area, ensuring that a complete underwater feature map is obtained;

[0051] S2: The industrial computer performs filtering: The collected feature image to be processed is subjected to image enhancement processing by using an algorithm combining adaptive contrast enhancement and limited contrast adaptive histogram equalization (CLAHE) to obtain an enhanced feature image. The specific process is as follows:

[0052] The feature image to be processed is divided into a high-frequency image and a low-frequency image through a template filtering operation, and the high-frequency image is further enhanced. The enhanced high-frequency image and the low-frequency image are fused through an image enhancement algorithm to obtain an enhanced feature image. The mathematical expression of the image enhancement algorithm is shown in the following formula: g(x, y) = m d (x,y)+G(x,y)[f(x,y)-m d (x,y)], where g(x,y) represents the grayscale value of the processed image at point (x,y), and f(x,y) represents the grayscale value of the image before processing at point (x,y); m d (x, y) is the low-frequency image part obtained after filtering, G(x, y) is the dynamic gain function, and the G(x, y) value can change with the change of the processed pixel points;

[0053] The mean filtering method is used as a means to obtain a low-frequency image. The specific mathematical expression is as follows: The ratio of the image grayscale mean and the standard deviation in the neighborhood is used as the dynamic gain, and the following formula is established: Where σ(x,y) is defined as follows: The template radius n is artificially set to 2, MJ represents the grayscale mean of the image, the coordinate is the standard deviation σ in the neighborhood of the point (x, y); ε is the correction parameter; and the coordinate is the dynamic gain G of the point (x, y).

[0054] S3: Denoising: Denoising the enhanced feature image using a denoising template to obtain a filtered optimized feature image;

[0055] Denoising template acquisition method: According to the characteristic that the morphological closing operation can fill the small holes in the enhanced feature image, the small holes in the enhanced feature image refer to the black spots composed of impurity noise information. In the enhanced feature image after the morphological closing operation, the number of black spots will be significantly reduced, and the originally black pixels will get new gray values; the pixel value of the black spots is significantly smaller than the average gray value of the image or the median gray value of the overall image; thus, the denoising template image can be obtained by the following formula: Among them, M(x,y) is the image element value of the denoising template, C(x,y) is the grayscale value of the image pixel after the closing operation, k is the noise judgment threshold, and k is a constant. The purpose of this operation is to compare the difference in pixel grayscale between the enhanced feature image and the enhanced feature image denoised by the closing operation, and mark the granular black spots in the enhanced feature image whose pixel grayscale difference between the enhanced feature image and the enhanced feature image denoised by the closing operation is greater than the preset threshold as potential noise points; the grayscale value of the potential noise point is replaced by the grayscale information of other points through the median filtering method, and the original enhanced feature image is updated after each noise filtering operation on a potential noise point, and finally a filtered optimized feature image is obtained, which can obtain a better filtering effect.

[0056] Embodiment 2:

[0057] S4: Build a deep learning model and set model training parameters, run the YOLOv8 network model to extract underwater weld feature information from the filtered optimized feature image;

[0058] Specifically, set the model training parameters and run the model to enable full-process monitoring of the training. The main things that need to be modified are the settings of the training rounds, batch size, and learning rate, and monitor the changes in the training set bounding box loss, target loss, and classification loss during the training process;

[0059] Furthermore, a data set is constructed based on the model training parameters, and the data set is divided into a training set and a validation set, wherein the training set is used for learning and adapting the model, and the validation set is used for monitoring and verifying the training process;

[0060] The training set is used to teach the neural network model, while the validation set is responsible for testing the effectiveness of the model learning results, helping to screen out the YOLOv8 network model with the best performance, thereby avoiding overfitting or underfitting problems;

[0061] The optimal YOLOv8 network model is used. The main structure is to divide the grid first, each grid is responsible for the prediction of the target in the corresponding area, and finally output after non-maximum suppression;

[0062] In this embodiment, for underwater feature detection, the number of detection categories is 1, the network model needs to configure the size of the prior box, the configuration of the network backbone Backbone and the configuration of the Head, and the relevant parameters include which layer the input comes from (from), the number (number), the module type (module) and other parameters (args). When building the network model, the depth coefficient depth_multipl e is multiplied by the number to adjust the depth, and the width coefficient width_multipl e is multiplied by the number of convolution kernels to obtain the actual number of convolution kernels. In this way, the width of the network model is adjusted;

[0063] Finally, the weld feature information is extracted through the optimal YOLOv8 target detection network;

[0064] S5: Accurately map the underwater weld feature information to the coordinate system of the underwater machine to calibrate the laser sensor posture;

[0065] Specifically, two coordinate systems are defined: {S} represents the sensor coordinate system, and {B} represents the reference coordinate system. The coordinates of point P in these two coordinate systems are expressed as B P (in the reference coordinate system) and S P (in the sensor coordinate system), the purpose of sensor posture calibration is to determine the transformation matrix of the sensor coordinate system {S} relative to the end coordinate system {E} as accurately as possible For any known point in its coordinate system S Q, the formula to transform it into the machine base coordinate system is:

[0066] Among them, the transformation matrix of the sensor coordinate system {S} relative to the end coordinate system {E} Calculate the transformation matrix The specific steps are as follows:

[0067] SS1: Determine a specific point P on the target and record the position of point P in the reference coordinate system {B} B P=(x B ,y B ,z B ,1) T ;

[0068] SS2: Adjust the position of the machine to ensure that the laser line emitted by the sensor can pass through point P. Also record the position of point P in the sensor coordinate system {S} S P=(x S ,0,z S ,1) T ;

[0069] SS3: Switch the current tool coordinate system of the machine to the end coordinate system {E}, record the pose data in the end coordinate system {E} at this time, and use Euler's rotation theorem to calculate the rotation matrix Further seek

[0070] SS4: According to the transformation relationship of point P in space:

[0071] For any known point in its coordinate system S Q, the formula to transform it into the machine base coordinate system is: In the formula, B Q is the position coordinate of point Q in the reference coordinate system {B}, S Q is the position coordinate of point Q in the sensor coordinate system {S}, is the calculation result of the formula, is the calculation result of the formula;

[0072] S6: After posture calibration, the RRT (rapid random tree) path planning algorithm is used to generate a feature tracking trajectory. The controller drives the welding robot to guide the welding gun to move according to the feature tracking trajectory. At the same time, the deviation warning feedback analysis is performed on the feature tracking trajectory of the welding robot, the obtained avoidance deviation index is discriminated, and the obtained adjustment instruction is output as feedback;

[0073] The real-time moving trajectory of the welding robot is obtained, and the real-time moving trajectory is compared with the generated feature tracking trajectory for coincidence, thereby obtaining the coincidence between the real-time moving trajectory and the generated feature tracking trajectory, and setting the coincidence between the real-time moving trajectory and the generated feature tracking trajectory as the moving deviation warning coefficient, and performing discrimination processing on the moving deviation warning coefficient:

[0074] If the movement deviation warning coefficient is less than the preset movement deviation warning coefficient threshold, a feedback instruction is generated;

[0075] If the movement deviation warning coefficient is greater than or equal to the preset movement deviation warning coefficient threshold, an alarm instruction is generated, a feedback instruction or an alarm instruction is output as feedback, and a preset warning operation corresponding to the feedback instruction or the alarm instruction is immediately performed, so as to timely adjust the real-time movement trajectory of the welding robot to improve the movement accuracy of the welding robot;

[0076] When the feedback instruction is generated, the generated feature tracking trajectory is divided into an avoidance tracking trajectory and a non-avoidance tracking trajectory, the real-time avoidance trajectory of the welding robot is obtained, the minimum avoidance distance of the avoidance tracking trajectory of the welding robot is obtained, and the minimum avoidance distance represents the minimum straight-line distance between the obstacle and the welding robot. At the same time, the actual minimum avoidance distance of the real-time avoidance trajectory of the welding robot is obtained, and the value obtained by subtracting the minimum avoidance distance from the actual minimum avoidance distance is set as the avoidance deviation index, and the avoidance deviation index is discriminated:

[0077] If the avoidance deviation index is greater than or equal to the preset avoidance deviation index threshold, no signal is generated;

[0078] If the avoidance deviation index is less than the preset avoidance deviation index threshold, an adjustment instruction is generated, the adjustment instruction is output as feedback, and the preset warning operation corresponding to the adjustment instruction is immediately performed, so as to timely manage and adjust the feature tracking trajectory generation plan to reduce the risk of obstacle avoidance failure of the welding robot;

[0079] In summary, the present invention analyzes from the perspective of image information processing to improve the image processing quality, and adopts an algorithm combining adaptive contrast enhancement with limited contrast adaptive histogram equalization to enhance image contrast, highlight feature edge information, and denoise the enhanced feature image by a denoising template method, effectively suppressing noise interference in underwater images, so as to provide data support for subsequent feature recognition, and extract underwater weld feature information from the filtered optimized feature image based on the YOLOv8 network model, thereby improving feature recognition precision and accuracy, and calibrating the laser sensor posture by feature information mapping to provide technical support for subsequent feature tracking, so as to improve feature tracking accuracy, and perform actual tracking self-checking on the generated feature tracking trajectory by information feedback, so as to timely adjust the real-time moving trajectory of the welding robot to improve the moving accuracy of the welding robot, and further manage and adjust the feature tracking trajectory generation scheme to reduce the risk of obstacle avoidance failure of the welding robot.

[0080] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by technicians in this field for each group of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0081] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A study on underwater feature recognition and tracking based on deep learning, characterized in that: The following steps are involved: S1, scanning underwater features, capturing underwater feature contour information, and setting the underwater feature contour information as a feature image to be processed; S2, the industrial computer performs filtering: the collected feature image to be processed is subjected to image enhancement processing by using an algorithm combining adaptive contrast enhancement and limited contrast adaptive histogram equalization (CLAHE) to obtain an enhanced feature image; S3, denoising: denoising the enhanced feature image using a denoising template to obtain a filtered optimized feature image; S4: Build a deep learning model and set model training parameters, run the YOLOv8 network model to extract underwater weld feature information from the filtered optimized feature image; S5: Accurately map the underwater weld feature information to the coordinate system of the underwater machine to calibrate the laser sensor posture; S6: After posture calibration, the RRT (rapid random tree) path planning algorithm is used to generate a feature tracking trajectory. The controller drives the welding robot to guide the movement of the welding gun according to the feature tracking trajectory, performs deviation warning feedback analysis on the feature tracking trajectory of the welding robot, performs discrimination processing on the obtained avoidance deviation index, and outputs the obtained adjustment instruction as feedback.

2. The method for underwater feature recognition and tracking based on deep learning according to claim 1 is characterized in that: The process of acquiring the enhanced feature image is as follows: The feature image to be processed is divided into a high-frequency image and a low-frequency image through a template filtering operation, and the high-frequency image is further enhanced. The enhanced high-frequency image and the low-frequency image are fused through an image enhancement algorithm to obtain an enhanced feature image. The mathematical expression of the image enhancement algorithm is shown in the following formula: g(x, y) = m d (x,y)+G(x,y)[f(x,y)-m d (x,y)], where g(x,y) represents the grayscale value of the processed image at point (x,y), and f(x,y) represents the grayscale value of the image before processing at point (x,y); m d (x, y) is the low-frequency image part obtained after filtering, G(x, y) is the dynamic gain function, and the G(x, y) value can change with the change of the processed pixel points; The mean filtering method is used as a means to obtain a low-frequency image. The specific mathematical expression is as follows: The ratio of the image grayscale mean and the standard deviation in the neighborhood is used as the dynamic gain, and the following formula is established: Among them, σ(x,y) is defined as follows: The template radius n is artificially set to 2.

3. The method for underwater feature recognition and tracking based on deep learning according to claim 2 is characterized in that: The denoising template acquisition method: The denoising template image is obtained by the following formula: Among them, M(x,y) is the image element value of the denoising template, C(x,y) is the grayscale value of the image pixel after the closing operation, k is the noise judgment threshold, and k is a constant. The purpose of this operation is to compare the difference in pixel grayscale between the enhanced feature image and the enhanced feature image denoised by the closing operation, and mark the granular black spots in the enhanced feature image whose pixel grayscale difference between the enhanced feature image and the enhanced feature image denoised by the closing operation is greater than the preset threshold as potential noise points; the grayscale value of the potential noise point is replaced by the grayscale information of other points through the median filtering method, and the original enhanced feature image is updated after each noise filtering operation on a potential noise point, and finally the filtered optimized feature image is obtained.

4. The method for underwater feature recognition and tracking based on deep learning according to claim 3 is characterized in that: The YOLOv8 network model construction process is as follows: Setting model training parameters, building a data set based on the model training parameters, and dividing the data set into a training set and a validation set. The training set is used to teach the neural network model, while the validation set is responsible for verifying the effectiveness of the model learning results. Run the model and start monitoring the entire training process. The main things that need to be modified are the settings of the training rounds, batch size, and learning rate. Monitor the changes in the training set bounding box loss, target loss, and classification loss during the training process, and ultimately help screen out the YOLOv8 network model with the best performance.

5. The method for underwater feature recognition and tracking based on deep learning according to claim 4 is characterized in that: The underwater weld feature information is accurately mapped to the coordinate system of the underwater machine in order to calibrate the laser sensor posture. The specific process is as follows: Define two coordinate systems: {S} represents the sensor coordinate system, {B} represents the reference coordinate system, and the coordinates of point P in these two coordinate systems are expressed as B P and S P; Determine the transformation matrix of the sensor coordinate system {S} relative to the end coordinate system {E} For any known point in its coordinate system S Q, the formula to transform it into the machine base coordinate system is:

6. The method for underwater feature recognition and tracking based on deep learning according to claim 5 is characterized in that: Transformation matrix of sensor coordinate system {S} relative to end coordinate system {E} Calculate the transformation matrix The specific steps are as follows: SS1: Determine a specific point P on the target and record the position of point P in the reference coordinate system {B} B P=(x B ,y B ,z B ,1) T ; SS2: Adjust the position of the machine to ensure that the laser line emitted by the sensor can pass through point P. Also record the position of point P in the sensor coordinate system {S} S P=(x S ,0,z S ,1) T ; SS3: Switch the current tool coordinate system of the machine to the end coordinate system {E}, record the pose data in the end coordinate system {E} at this time, and use Euler's rotation theorem to calculate the rotation matrix Further seek SS4: According to the transformation relationship of point P in space: For any known point in its coordinate system S Q, the formula to transform it into the machine base coordinate system is: In the formula, B Q is the position coordinate of point Q in the reference coordinate system {B}, S Q is the position coordinate of point Q in the sensor coordinate system {S}.

7. The method for underwater feature recognition and tracking based on deep learning according to claim 6 is characterized in that: The deviation warning feedback analysis process is as follows: The real-time moving trajectory of the welding robot is obtained, and the real-time moving trajectory is compared with the generated feature tracking trajectory for coincidence, thereby obtaining the coincidence between the real-time moving trajectory and the generated feature tracking trajectory, and setting the coincidence between the real-time moving trajectory and the generated feature tracking trajectory as the moving deviation warning coefficient, and performing discrimination processing on the moving deviation warning coefficient to generate a feedback instruction or generate an alarm instruction; When generating a feedback instruction, the generated feature tracking trajectory is divided into an avoidance tracking trajectory and a non-avoidance tracking trajectory, the real-time avoidance trajectory of the welding robot is obtained, the minimum avoidance distance of the avoidance tracking trajectory of the welding robot is obtained, the minimum avoidance distance represents the minimum straight-line distance between the obstacle and the welding robot, and the actual minimum avoidance distance of the real-time avoidance trajectory of the welding robot is obtained. The value obtained by subtracting the minimum avoidance distance from the actual minimum avoidance distance is set as the avoidance deviation index, and the avoidance deviation index is judged: if the avoidance deviation index is greater than or equal to the preset avoidance deviation index threshold, no signal is generated; if the avoidance deviation index is less than the preset avoidance deviation index threshold, an adjustment instruction is generated.

Images