Spraying robot operation path planning method based on YOLOv5 improved artificial potential field method

Abstract

The invention discloses a spraying robot operation path planning method based on a YOLOv5 improved artificial potential field method. A machine body structure, a power system, a spraying end execution device, a control system, an information acquisition system, a safety system and a robot algorithm system are included. The machine body structure comprises a base, a lifting system, a support arm, a joint and a control panel; the power system comprises a motor and a pneumatic and hydraulic module; the information acquisition system comprises a sensor and a camera module; the spraying end execution device comprises a spraying gun and a nozzle; the safety system comprises an emergency shutdown module and a sensor protection and anti-falling module; the robot algorithm system comprises a YOLOv5 vision algorithm and an artificial potential field method for path planning; the YOLOv5 vision algorithm system comprises a model architecture, training data, a loss function, an optimizer, an inference engine and the like; the artificial potential field method path planning system comprises potential field generation, a path planning algorithm and obstacle detection. Compared with traditional manual work, the working efficiency can be improved by 10%, and 3-4 workers are saved in long-term use.

Description

Technical Field

[0001] The invention relates to a spraying robot operation path planning method based on a YOLOv5 improved artificial potential field method, and belongs to the technical field of shipbuilding engineering in ship and ocean engineering. Background Art

[0002] In order to develop a new model of intelligent manufacturing and accelerate the transformation to digitalization and digital intelligence, the development of various fields tends to intelligent manufacturing. The same is true for the technological development in the field of ships. Similarly, the path planning of high-altitude spraying robots has emerged. The path planning technology of the execution end of the high-altitude steel outer surface spraying of ships is to create a completely autonomous, digital, and intelligent spraying method through artificial intelligence. In the context of spraying technology, intelligent technology can be used to create an intelligent high-altitude spraying robot that can achieve efficient, accurate, and safe spraying in high-altitude operations, greatly improving the efficiency and quality of traditional spraying operations. At the same time, it also reduces the risks and labor intensity of manual labor, providing strong support for high-altitude operations in modern industry.

[0003] At present, in the era of big data, Internet of Things, and artificial intelligence, the use of intelligent high-altitude spraying robot technology to achieve automatic spraying and autonomous planning of spraying trajectories can shorten the operation time, improve process accuracy, and increase productivity; reduce the risks of manual work in high-altitude operations, protect the safety of workers, reduce manpower requirements and material waste, thereby reducing overall operating costs, and be able to operate in harsh or difficult-to-reach environments. This technology has become an inevitable trend. In this context, we must carry out digital reforms on traditional shipyards, that is, to make an intelligent high-altitude spraying robot that uses intelligent algorithm technology to plan the trajectory of spraying work. At the same time, according to national policies, we must achieve low-carbon, green, environmentally friendly, quick and easy manufacturing methods. Nowadays, the working environment of shipyards is harsh, the spraying work is seriously polluted, the noise is loud, the water and electricity waste, and the efficiency is low. These problems cannot be alleviated. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a spray robot operation path planning method based on the YOLOv5 improved artificial potential field method, thereby solving the above-mentioned technical problems.

[0005] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a method for planning the operation path of a spraying robot based on the improved artificial potential field method of YOLOv5, including a body structure, a power system, a spraying end effector, a control system, an information collection system, a safety system and a robot algorithm system;

[0006] The body structure is used to stabilize the body, maintain balance during high-altitude operations, and ensure that the robot does not tilt during spraying; it supports the movement of the robot at different heights and angles so that the robot can cover various parts of the spraying area;

[0007] The power system is used to provide power to drive the robot's walking, lifting, rotation and other functions, provide the necessary pressure and flow for the spraying system, and provide the required electricity, hydraulic, pneumatic and other energy for the entire system to ensure the normal operation of all components and functions;

[0008] The spraying end effector is used to complete the spraying of a small area unit of 2m*3m. The servo arm vehicle drives the spraying end effector to move on the outer surface of the hull body to achieve coverage of the entire spraying area of ​​the hull body plate. The sensor detects the distance between the four corner points of the spraying end effector and the hull body plate. According to the four distance values, the servo arm vehicle automatically moves to adjust the distance and relative angle between the spraying end effector and the wall surface. The spraying end effector includes a distance sensor, a slide, a robotic arm, a frame, a spray gun clamping mechanism, etc.

[0009] The control system is used to accurately control the movement of the robot, adjust the operating parameters of the spraying equipment, formulate and execute the spraying path planning, process data from various sensors, monitor the robot status and environment, record the data during the spraying process, etc., so as to meet the task requirements of high-altitude spraying; distribute the processed information and feed the data back to the YOLOv5 visual algorithm system and the artificial potential field method path planning system for the subsequent path planning; the control system includes a sensor module, a visual processing module, a path planning module, a control algorithm module, an actuator module, a communication module, and a user interface module.

[0010] The information collection system is used to monitor the surrounding environment in real time, and accurately locate the received environmental data through a visual system or a laser sensor to help the robot complete path planning;

[0011] The robot algorithm system is a YOLOv5 visual algorithm, which is used to identify and locate spraying areas, obstacles, areas to be sprayed, and over-spraying areas; the identified image is transmitted to the control system through a visual camera, and the control system processes the data and feeds it back to the YOLOv5 visual algorithm system; after being optimized by the YOLOv5 visual algorithm system, the artificial potential field algorithm is assisted to complete path planning; the algorithm includes target detection, feature extraction, bounding box regression, classification and confidence scoring, real-time processing, multi-scale detection, training and fine-tuning, and is used to detect image features of ship outer plates;

[0012] The artificial potential field algorithm is used to generate virtual attraction and repulsion using the artificial potential field method to help the end effector avoid obstacles and move toward the target point; the artificial potential field method cooperates with the YOLOv5 visual algorithm to adjust the movement direction of the robot in real time to ensure the smooth progress of the spraying process; when processing multiple targets or areas, the artificial potential field method can help the robot coordinate between different targets to achieve multi-point spraying tasks;

[0013] The specific method is as follows: first, the spraying area is divided into two upper and lower sections, and two multi-degree-of-freedom robotic arms are equipped with automatic spray guns to realize spraying; a synchronous belt drive is used to achieve a higher moving speed and start-stop acceleration, and the robotic arm is equipped to complete horizontal movement. The spray gun muzzle is arranged vertically, and the area spraying of the corresponding height is completed through multiple horizontal movements; the YOLOv5 visual algorithm is used to ensure the position and surrounding environment of the spray gun, and the specific position and surrounding characteristics obtained are fed back to the YOLOv5 visual algorithm system, and the information is integrated and transmitted to the artificial potential field method path planning system; according to the gravitational field function and repulsive field function of the artificial potential field method, the non-steel outer surface and non-paint color identified by the YOLOv5 visual algorithm are calculated as obstacles, so that the artificial potential field method is used to avoid obstacles and obtain an ideal path planning route.

[0014] Furthermore, the YOLOv5 visual algorithm includes target detection, feature extraction, bounding box regression, classification and confidence scoring, real-time processing, multi-scale detection, training and fine-tuning, and is used to detect image features of ship outer plates; the algorithm is specifically:

[0015] First, a two-dimensional Gaussian distribution model is built for the image bounding box, so that the value at the center of the bounding box is the highest, and then gradually decreases to both sides, so that the convergence and robustness of the image are gradually improved; then the difference operation is performed on the average vector value of the Gaussian distribution, and finally the cosine function is used to normalize it in the interval [π, 2π] to obtain the improved bounding box loss function;

[0016] The bounding box loss function of the YOLOv5 visual recognition algorithm is improved by the following formula; the probability density function of the two-dimensional Gaussian distribution is:

[0017]

[0018] Where x, μ and ∑ are the pixel center coordinates, mean vector and covariance matrix of Gaussian distribution respectively; the improved bounding box loss function obtained by the second-order definition function of Gaussian distribution function μ is:

[0019]

[0020] Where: w is the width of the predicted bounding box; c is the diagonal distance of the minimum closed area containing the predicted box and the true box; Np is the Gaussian distribution model of the prediction box P; N g is the Gaussian distribution model for annotating the ground-truth box.

[0021] Furthermore, the artificial potential field method path planning includes potential field generation, target attraction, obstacle repulsion, synthetic potential field, path optimization, and dynamic adjustment; the distance influence factor is used to improve the gravity function to solve the problem that the underwater robot is prone to collision with obstacles when the underwater robot is far away from the target point; the improved gravity function is:

[0022]

[0023] Where K att Gravity gain coefficient, q represents the current position of the robot, q g Represents the location of the target point, and d is a certain set distance. Through the above improvements, the improved artificial potential field method can be used for effective spraying according to the obstacle target defined by the YOLOv5 visual recognition algorithm.

[0024] Furthermore, the information is fed back to the robot main device through the information acquisition system. After reading the feedback data, the control system of the robot main device issues an execution command, and the robot completes the real-time command.

[0025] Furthermore, firstly, the visual sensor collects data and images, and then changes the image scale of the collected images to improve the robustness and accuracy of the target detection model. Firstly, the visual sensor collects data and images, and then changes the image scale of the collected images to improve the robustness and accuracy of the target detection model.

[0026] During the training process, YOLOv5 uses data augmentation technology to handle image scale changes. In each training iteration, YOLOv5 randomly adjusts the size of the input image to ensure that the model can be trained on images of different sizes, thereby improving the detection ability of different object sizes.

[0027] In order to maintain the aspect ratio of the input image, YOLOv5 will pad and crop the image. In the inference stage, the YOLOv5 network model usually requires the size of the input image to match the size during model training. Secondly, the environment and weights are configured to create a virtual spraying environment, and the collected image information is used for pre-training the model.

[0028] Then the process of training the YOLOv5 image information model includes data preparation, model configuration, training, evaluation and testing. Ensure that the data set is annotated in the YOLO5 format and the model file and training parameters are correctly configured to effectively train and evaluate the object detection model.

[0029] The optimal configuration weight after training is selected, and then visual judgment is performed to determine whether it is a steel outer surface or a painted surface. If the judgment result is a steel outer surface or a painted surface, the next step of calculation is started. If not, the data collection stage is returned. Finally, the correct result is inferred and predicted. After inference and prediction, the object features and color analysis are used to pave the way for the next step of path planning.

[0030] Furthermore, in order to evaluate the accuracy of prediction in the target detection model, the mean average precision (mAP) and the frame rate (FPS) per second (FPS) are used as the basis for evaluating the prediction effect; wherein, the mAP calculation formula is shown in formulas (1) to (4):

[0031]

[0032] Among them, TP represents the number of detection boxes, AP represents the accuracy value of each category, N is the number of target types, p represents precision, r represents recall rate, and i represents each category; the parameter FPS is used to evaluate the running speed of the model. The larger the FPS, the faster the recognition speed of the current system.

[0033] Furthermore, the specific implementation method of the YOLOv5 network model is as follows: by inputting the surface features of the ship, using Backbone to extract the image features, connecting Backbone and Head to further increase the image features, Head is responsible for predicting the bounding box and category, outputting the detection results, and calculating the model loss function through Loss to further analyze the color results.

[0034] Furthermore, the artificial potential field method path planning includes potential field generation, target attraction, obstacle repulsion, synthetic potential field, path optimization, and dynamic adjustment; the on-site environment and dynamic characteristics of the obstacle are defined according to the YOLOv5 visual algorithm, and the defined obstacle information and characteristic function are transmitted to the artificial potential field method; the gravitational potential field function calculated by the artificial potential field method is:

[0035]

[0036] The gravitational function is:

[0037]

[0038] Decompose gravity into two directions: x and y

[0039] F attx (q)=-K att (xx goal ) (7);

[0040] Fatty (q)=-K att (yy goal ) (8);

[0041] Where K att is the scale factor, ρ(q,q goal ) represents the deviation between the robot’s current position and the target position;

[0042] The repulsive potential field function is:

[0043]

[0044] Where K rep is the repulsive force scale factor, ρ(q,q obs ) represents the spatial distance from the robot to the obstacle, ρ 0 represents the influence range of the obstacle. Similarly, the repulsive force is the repulsive field gradient;

[0045] Repulsion function:

[0046]

[0047] Decomposing the repulsive force into components in the x and y directions is:

[0048]

[0049] The total field is the sum of the repulsive field and the gravitational field.

[0050] U 合 =U att +U rep (13);

[0051] The total force is also the accumulation of repulsion and attraction (in x and y directions)

[0052] F x合 =F attx +F repx (14);

[0053] F y合 =F atty +F repy (15);

[0054] Finally, multiply the step length to reach the new position and then iterate;

[0055] The specific method of the above function is: first, the gravitational function is used to calculate the obstacle target to determine the distance between the obstacle and the spray gun; secondly, the repulsive function is used to calculate the obstacle target so that the force of the obstacle produces a repulsive force on the spray gun; finally, the resultant force of the two is calculated to obtain a suitable position; if it is not zero, proceed to the next target point; if the resultant force is zero, the simulated annealing algorithm is integrated to add the target point for optimization until the resultant force is not zero, and then proceed to the next target point; according to the artificial potential field method, it is determined whether the target point is reached accurately. If the target point is reached accurately, the algorithm ends, otherwise the algorithm program continues to cycle from the calculation of the resultant force.

[0056] Furthermore, a fully intelligent and digital high-altitude steel surface spraying robot end effector was obtained by combining the YOLOv5 visual algorithm with the artificial potential field method path planning.

[0057] The working principle of the robot end effector is as follows: first, the robot arm approaches the end spraying surface; second, the visual algorithm is started to define the external features of the obstacle and the surrounding environment. The visual sensor is used to collect image information and fed back to the YOLOv5 visual algorithm to determine whether it is an obstacle. If it is an obstacle, the next step is taken. Otherwise, it returns to the image collection stage and repeats the above actions; then the judged information is fed back to the control system, so that the artificial potential field method starts to calculate; the starting point and the target point are set, and the path planning starts to find the way; whether to start the obstacle avoidance operator after encountering an obstacle; normal obstacle avoidance means driving to the target point and ending; if the obstacle avoidance is invalid, it returns to the control system to continue judging.

[0058] The beneficial effects of the present invention are as follows: the present invention proposes a method for planning the operation path of a spraying robot based on the improved artificial potential field method of YOLOv5, wherein the robot includes a body structure, a power system, a spraying terminal execution device, a control system, an information acquisition system, a safety system, and a robot algorithm system, and the information acquisition system is used to feed back to the robot body device, and after reading the feedback data, the control system of the robot body device issues an execution command, and the robot completes the real-time command. The information recognition, collection, analysis, decision-making, path planning, obstacle avoidance and other functions of the intelligent spraying execution end path planning of the high-altitude steel outer surface of the ship are realized to improve the work efficiency of the factory, monitor the working information of the equipment, reduce accidents caused by production, and reduce the waste of resources and environmental pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The robot operation process flow chart of the present invention;

[0060] Figure 2 is a flow chart of the robot system of the present invention;

[0061] Figure 3 It is the YOLOv5 visual algorithm flow chart of the present invention;

[0062] Figure 4 It is a YOLOv5 network model framework diagram of the present invention;

[0063] Figure 5 It is a YOLOv5 network model framework diagram of the present invention;

[0064] Figure 6 This is a control flow chart of the spraying end path planning of the present invention;

[0065] Figure 7 This is a control flow chart of the spraying end path planning of the present invention;

[0066] Figure 8 This is a comparison diagram of execution end path planning before and after the improvement of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below through the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0069] like Figure 1 As shown, the present invention provides a method for planning the operation path of a spraying robot based on the improved artificial potential field method of YOLOv5, including a body structure, a power system, a spraying terminal execution device, a control system, an information acquisition system, a safety system, and a robot algorithm system; the present invention divides the 2x3m unit spraying area into two upper and lower intervals, and uses two multi-degree-of-freedom mechanical arms to carry automatic spray guns to achieve spraying. The X-axis horizontal slide has a moving stroke of 3m, and a synchronous belt drive is used to achieve a higher moving speed and start-stop acceleration. The mechanical arm is equipped to complete horizontal movement, and the spray gun muzzle is arranged vertically. The spraying of an area with a height of 1m is completed through 3-4 horizontal movements. The position and surrounding environment of the spray gun are guaranteed by the YOLOv5 visual algorithm, and the specific position and surrounding characteristics obtained are fed back to the YOLOv5 visual algorithm system, and the information is integrated and transmitted to the artificial potential field method path planning system. According to the gravitational field function and repulsive field function of the artificial potential field method, the non-steel outer surface and non-painted color identified by the YOLOv5 visual algorithm are calculated as obstacles, and the artificial potential field method is used to perform obstacle avoidance processing to obtain the ideal path planning route.

[0070] The specific implementation method of the robot system flow chart in the present invention is as follows Figure 2 As shown, the information is fed back to the robot main device through the information acquisition system. After reading the feedback data, the control system of the robot main device issues an execution command, and the robot completes the real-time command.

[0071] The specific implementation method of the YOLOv5 visual algorithm in the present invention is as follows Figure 3 As shown in the figure, firstly, the visual sensor collects data and images, and the collected images are scaled to improve the robustness and accuracy of the target detection model. Image scale change involves scaling the images during the training and inference stages to ensure that the model can handle target objects of different sizes. During the training process, YOLOv5 uses data augmentation technology to handle image scale changes. In each training iteration, YOLOv5 randomly adjusts the size of the input image. Ensure that the model can be trained on images of different sizes, thereby improving the detection ability of different target sizes; in order to maintain the aspect ratio of the input image, YOLOv5 will pad and crop the image; the YOLOv5 model usually requires the size of the input image to match the size of the model during training during the inference stage. Secondly, configure the environment and weights, create a virtual spraying environment, and pre-train the model with the collected image information. Then, the process of training the YOLOv5 image information model includes data preparation, model configuration, training, evaluation and testing. Ensure that the data set is labeled in the YOLO5 format, correctly configure the model file and training parameters, and effectively train and evaluate the target detection model. The optimal configuration weight after training is selected, and then visual judgment is performed to determine whether it is a steel outer surface or a painted surface. If the judgment result is a steel outer surface or a painted surface, the next step of calculation is started. If not, the data collection stage is returned. Finally, the correct result is inferred and predicted. After inference and prediction, the object features and color analysis are used to pave the way for the next step of path planning.

[0072] In order to evaluate the accuracy of the prediction, the average precision value mAP (mean Average Precision) and frame rate per second FPS (Frame Per Second) are selected as the basis for evaluating the prediction effect.

[0073] The mAP calculation formula is shown in formulas (1) to (4):

[0074]

[0075] Among them, TP represents the number of detection boxes, AP represents the accuracy value of each category, N is the number of target types, p represents precision, r represents recall, and i represents each category. The parameter FPS is used to evaluate the running speed of the model. The larger the FPS, the faster the recognition speed of the current system.

[0076] The specific implementation method of the YOLOv5 network model framework in the present invention is as follows Figure 4 As shown in the figure, by inputting the surface features of the ship, Backbone is used to extract the image features, and Backbone and Head are connected to further enhance the image features. Head is responsible for predicting the bounding box and category, outputting the detection results, and calculating the model loss function through Loss to further analyze the color results.

[0077] The specific implementation method of the artificial potential field path planning in the present invention is as follows Figure 5 As shown in the figure, the artificial potential field method path planning includes potential field generation, target attraction, obstacle repulsion, synthetic potential field, path optimization, and dynamic adjustment; the on-site environment and dynamic characteristics of the obstacle are defined according to the YOLOv5 visual algorithm, and the defined obstacle information and characteristic functions are transmitted to the artificial potential field method. The gravitational potential field function calculated by the artificial potential field method is:

[0078]

[0079] The gravitational function is:

[0080]

[0081] Decompose gravity into two directions: x and y

[0082] F attx (q)=-K att (xx goal ) (7)

[0083] F atty (q)=-K att (yy goal ) (8)

[0084] Where K att is the scale factor, ρ(q,q goal ) represents the deviation between the robot’s current position and the target position.

[0085] The repulsive potential field function is:

[0086]

[0087] Where K rep is the repulsive force scale factor, ρ(q,q obs ) represents the spatial distance from the robot to the obstacle, ρ 0 It represents the influence range of the obstacle. Similarly, the repulsive force is the repulsive field gradient.

[0088] Repulsion function:

[0089]

[0090] Decomposing the repulsive force into components in the x and y directions is:

[0091]

[0092] The total field is the sum of the repulsive field and the gravitational field.

[0093] U 合 =U att +U rep (13) The total force is the sum of repulsion and attraction (in the x and y directions).

[0094] F x合 =F attx +F repx (14)

[0095] F y合 =F atty +F repy (15) Finally, multiply the step length to reach the new position and then iterate. First, use the gravitational function to calculate the obstacle target and determine the distance between the obstacle and the spray gun; secondly, use the repulsive function to calculate the obstacle target so that the force of the obstacle produces a repulsive force on the spray gun; finally, calculate the resultant force of the two to get the appropriate position. If it is not zero, move to the next target point. If the resultant force is zero, add the target point for optimization by integrating the simulated annealing algorithm until the resultant force is not zero, and then move to the next target point. Determine whether the target point is reached accurately according to the artificial potential field method. If the target point is reached accurately, the algorithm ends. Otherwise, the algorithm program continues to cycle from the calculation of the resultant force.

[0096] The specific implementation method of the spraying end path planning in the present invention is as follows Figure 6 As shown in the figure, a fully intelligent and digital high-altitude steel surface spraying robot end effector is obtained by combining the YOLOv5 visual algorithm with the artificial potential field method path planning. First, the robot arm approaches the end spraying surface; secondly, the visual algorithm is started to define the external features and surrounding environment of the obstacle. The visual sensor is used to collect image information and fed back to the YOLOv5 visual algorithm to determine whether it is an obstacle. If it is an obstacle, the next step is taken. Otherwise, it returns to the image collection stage to repeat the above actions; then the judged information is fed back to the control system, so that the artificial market method starts to calculate; the starting point and the target point are set, and the path planning starts to find the way; whether to start the obstacle avoidance operator after encountering an obstacle. Normal obstacle avoidance ends at the target point. If the obstacle avoidance is invalid, it returns to the control system to continue to judge. Compared with traditional manual work, it can improve work efficiency by 10%, and long-term use can save 3 to 4 people's labor.

[0097] The specific implementation method of the improved YOLOv5 visual recognition algorithm in the present invention is as follows Figure 7As shown, this is to avoid the problems of low positioning accuracy, poor robustness, and difficulty in capturing target shape of the traditional YOLOv5 visual recognition algorithm.

[0098] The specific implementation method of improving the front and rear execution end path planning in the present invention is as follows Figure 8 As shown in the figure, the path planning based on the traditional artificial potential field method is prone to collision with obstacles and falling into extreme values. The distance influence factor is used to improve the gravity function to solve the problem that the underwater robot is prone to collision with obstacles when it is far away from the target point. Through the above improvements, the improved artificial potential field method can be used for effective spraying according to the obstacle target defined by the YOLOv5 visual recognition algorithm.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent substitution or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A spray robot operation path planning method based on YOLOv5 improved artificial potential field method, characterized in that: It includes the body structure, power system, spraying end-effector, control system, information collection system, safety system and robot algorithm system in the spraying end-effector loaded by the servo arm vehicle and the end of the arm vehicle; The body structure is used to stabilize the body, maintain balance during high-altitude operations, and ensure that the robot does not tilt during spraying; it supports the movement of the robot at different heights and angles so that the robot can cover various parts of the spraying area; The power system is used to provide power to drive the robot's walking, lifting, rotation and other functions, provide the necessary pressure and flow for the spraying system, and provide the required electricity, hydraulic, pneumatic and other energy for the entire system to ensure the normal operation of all components and functions; The spraying end effector is used to complete the spraying of a small area unit of 2m*3m. The servo arm vehicle drives the spraying end effector to move on the outer surface of the hull body to achieve coverage of the entire spraying area of ​​the hull body plate. The sensor detects the distance between the four corner points of the spraying end effector and the hull body plate. The servo arm vehicle automatically moves to adjust the distance and relative angle between the spraying end effector and the wall according to the four distance values. The control system is used to accurately control the movement of the robot, adjust the operating parameters of the spraying equipment, formulate and execute the spraying path planning, process data from various sensors, monitor the robot status and environment, record the data during the spraying process, etc., so as to meet the task requirements of high-altitude spraying; distribute the processed information and feed the data back to the YOLOv5 visual algorithm system and the artificial potential field method path planning system for the production of the subsequent path planning; The information collection system is used to monitor the surrounding environment in real time, and accurately locate the received environmental data through a visual system or a laser sensor to help the robot complete path planning; The robot algorithm system is a YOLOv5 visual algorithm, which is used to identify and locate spraying areas, obstacles, areas to be sprayed, and over-spraying areas; the identified image is transmitted to the control system through a visual camera, and the control system processes the data and feeds it back to the YOLOv5 visual algorithm system; after being optimized by the YOLOv5 visual algorithm system, the artificial potential field algorithm is assisted to complete path planning; the algorithm includes target detection, feature extraction, bounding box regression, classification and confidence scoring, real-time processing, multi-scale detection, training and fine-tuning, and is used to detect image features of ship outer plates; The artificial potential field algorithm is used to generate virtual attraction and repulsion using the artificial potential field method to help the end effector avoid obstacles and move toward the target point; the artificial potential field method cooperates with the YOLOv5 visual algorithm to adjust the movement direction of the robot in real time to ensure the smooth progress of the spraying process; when processing multiple targets or areas, the artificial potential field method can help the robot coordinate between different targets to achieve multi-point spraying tasks; The specific method is as follows: first, the spraying area is divided into two upper and lower sections, and two multi-degree-of-freedom robotic arms are equipped with automatic spray guns to realize spraying; a synchronous belt drive is used to achieve a higher moving speed and start-stop acceleration, and the robotic arm is equipped to complete horizontal movement. The spray gun muzzle is arranged vertically, and the area spraying of the corresponding height is completed through multiple horizontal movements; the YOLOv5 visual algorithm is used to ensure the position and surrounding environment of the spray gun, and the specific position and surrounding characteristics obtained are fed back to the YOLOv5 visual algorithm system, and the information is integrated and transmitted to the artificial potential field method path planning system; according to the gravitational field function and repulsive field function of the artificial potential field method, the non-steel outer surface and non-paint color identified by the YOLOv5 visual algorithm are calculated as obstacles, so that the artificial potential field method is used to avoid obstacles and obtain an ideal path planning route.

2. According to claim 1, a spraying robot operation path planning method based on YOLOv5 improved artificial potential field method is characterized in that: The YOLOv5 visual algorithm includes target detection, feature extraction, bounding box regression, classification and confidence scoring, real-time processing, multi-scale detection, training and fine-tuning, and is used to detect image features of ship outer plates; the specific algorithm is: First, a two-dimensional Gaussian distribution model is built for the image bounding box, so that the value at the center of the bounding box is the highest, and then gradually decreases to both sides, so that the convergence and robustness of the image are gradually improved; then the difference operation is performed on the average vector value of the Gaussian distribution, and finally the cosine function is used to normalize it in the interval [π, 2π] to obtain the improved bounding box loss function; The bounding box loss function of the YOLOv5 visual recognition algorithm is improved by the following formula; the probability density function of the two-dimensional Gaussian distribution is: Where x, μ and ∑ are the pixel center coordinates, mean vector and covariance matrix of Gaussian distribution respectively; the improved bounding box loss function obtained by the second-order definition function of Gaussian distribution function μ is: Where: w is the width of the predicted bounding box; c is the diagonal distance of the minimum closed area containing the predicted box and the true box; N p is the Gaussian distribution model of the prediction box P; N g is the Gaussian distribution model for annotating the ground-truth box.

3. According to claim 1, a spraying robot operation path planning method based on YOLOv5 improved artificial potential field method is characterized in that: The artificial potential field method path planning includes potential field generation, target attraction, obstacle repulsion, synthetic potential field, path optimization, and dynamic adjustment. The distance influence factor is used to improve the gravity function to solve the problem that the underwater robot is prone to collision with obstacles when the robot is far away from the target point. The improved gravity function is: Where K att Gravity gain coefficient, q represents the current position of the robot, q g Represents the location of the target point, and d is a certain set distance. Through the above improvements, the improved artificial potential field method can be used for effective spraying according to the obstacle target defined by the YOLOv5 visual recognition algorithm.

4. According to claim 1, a spraying robot operation path planning method based on YOLOv5 improved artificial potential field method is characterized in that: The information is fed back to the robot device through the information acquisition system. After reading the feedback data, the control system of the robot device issues an execution command, and the robot completes the real-time command.

5. According to claim 4, a spraying robot operation path planning method based on YOLOv5 improved artificial potential field method is characterized in that: First, the visual sensor collects data and images, and then changes the image scale of the collected images to improve the robustness and accuracy of the target detection model. First, the visual sensor collects data and images, and then changes the image scale of the collected images to improve the robustness and accuracy of the target detection model. During the training process, YOLOv5 uses data augmentation technology to handle image scale changes. In each training iteration, YOLOv5 randomly adjusts the size of the input image to ensure that the model can be trained on images of different sizes, thereby improving the detection ability of different object sizes. In order to maintain the aspect ratio of the input image, YOLOv5 will pad and crop the image. In the inference stage, the YOLOv5 network model usually requires the size of the input image to match the size during model training. Secondly, the environment and weights are configured to create a virtual spraying environment, and the collected image information is used for pre-training the model. Then the process of training the YOLOv5 image information model includes data preparation, model configuration, training, evaluation and testing. Ensure that the data set is annotated in the YOLO5 format and the model file and training parameters are correctly configured to effectively train and evaluate the object detection model. The optimal configuration weight after training is selected, and then visual judgment is performed to determine whether it is a steel outer surface or a painted surface. If the judgment result is a steel outer surface or a painted surface, the next step of calculation is started. If not, the data collection stage is returned. Finally, the correct result is inferred and predicted. After inference and prediction, the object features and color analysis are used to pave the way for the next step of path planning.

6. According to claim 5, a method for spraying robot operation path planning based on YOLOv5 improved artificial potential field method is characterized in that: In order to evaluate the accuracy of prediction in the target detection model, the mean average precision (mAP) and frame rate (FPS) are used as the basis for evaluating the prediction effect. The mAP calculation formula is shown in formulas (1) to (4): Among them, TP represents the number of detection boxes, AP represents the accuracy value of each category, N is the number of target types, p represents precision, r represents recall rate, and i represents each category; the parameter FPS is used to evaluate the running speed of the model. The larger the FPS, the faster the recognition speed of the current system.

7. The method for spraying robot operation path planning based on the YOLOv5 improved artificial potential field method according to claim 6 is characterized in that: The specific implementation method of the YOLOv5 network model is as follows: by inputting the surface features of the ship, using Backbone to extract the image features, connecting Backbone and Head to further increase the image features, Head is responsible for predicting the bounding box and category, outputting the detection results, and calculating the model loss function through Loss to further analyze the color results.

8. According to claim 4, a spraying robot operation path planning method based on YOLOv5 improved artificial potential field method is characterized in that: The artificial potential field method path planning includes potential field generation, target attraction, obstacle repulsion, synthetic potential field, path optimization, and dynamic adjustment; the on-site environment and dynamic characteristics of the obstacle are defined according to the YOLOv5 visual algorithm, and the defined obstacle information and characteristic function are transmitted to the artificial potential field method; the gravitational potential field function calculated by the artificial potential field method is: The gravitational function is: Decompose gravity into two directions: x and y F attx (q)=-K att (x-x goal ) (7); F atty (q)=-K att (y-y goal ) (8); Where K att is the scale factor, ρ(q,q goal ) represents the deviation between the robot’s current position and the target position; The repulsive potential field function is: Where K rep is the repulsive force scale factor, ρ(q,q obs ) represents the spatial distance from the robot to the obstacle, ρ0 represents the influence range of the obstacle, and similarly, the repulsive force is the repulsive field gradient; Repulsion function: Decomposing the repulsive force into components in the x and y directions is: The total field is the sum of the repulsive field and the gravitational field. IN 合 =U att +U rep (13); The total force is also the accumulation of repulsion and attraction (in x and y directions) F x合 =F attx +F repx (14); F y合 =F atty +F repy (15); Finally, multiply the step length to reach the new position and then iterate; The specific method of the above function is: first, the gravitational function is used to calculate the obstacle target to determine the distance between the obstacle and the spray gun; secondly, the repulsive function is used to calculate the obstacle target so that the force of the obstacle produces a repulsive force on the spray gun; finally, the resultant force of the two is calculated to obtain a suitable position; if the resultant force is not zero, proceed to the next target point; if the resultant force is zero, the simulated annealing algorithm is integrated to add the target point for optimization until the resultant force is not zero, and then proceed to the next target point; The artificial potential field method is used to determine whether the target point has been reached accurately. If the target point has been reached accurately, the algorithm ends. Otherwise, the algorithm program continues to cycle from the calculation of the resultant force.

9. The method for spraying robot operation path planning based on the YOLOv5 improved artificial potential field method according to claim 1, characterized in that: By combining the YOLOv5 visual algorithm with the artificial potential field method path planning, a fully intelligent and digital high-altitude steel surface spraying robot end effector was obtained; The working principle of the robot end effector is as follows: first, the robot arm approaches the end spraying surface; second, the visual algorithm is started to define the external features of the obstacle and the surrounding environment. The visual sensor is used to collect image information and fed back to the YOLOv5 visual algorithm to determine whether it is an obstacle. If it is an obstacle, the next step is taken. Otherwise, it returns to the image collection stage and repeats the above actions; then the judged information is fed back to the control system, so that the artificial potential field method starts to calculate; the starting point and the target point are set, and the path planning starts to find the way; whether to start the obstacle avoidance operator after encountering an obstacle; normal obstacle avoidance means driving to the target point and ending; if the obstacle avoidance is invalid, it returns to the control system to continue judging.

Images