Inspection path determination method and device based on self-labeled learning, equipment and medium

Abstract

The application discloses a method and device for determining an inspection path based on self-labeled learning, equipment and a medium. The method comprises: obtaining an initial inspection path output by an inspection autonomous decision model; controlling an inspection robot to move according to the initial inspection path; in the case that the effective perception time length proportion of an inspection perception object corresponding to the initial inspection path does not reach a preset proportion threshold, determining a target gradual degree of each path section in the initial inspection path to determine a deviation path section, and constructing target state constraint information corresponding to the deviation path section; generating an optimal inspection path according to the target state constraint information, and adding the optimal inspection path to a training data set of the inspection autonomous decision model for training and updating the inspection autonomous decision model. The technical solution guarantees the continuous monitorability of the inspection perception object in the movement process of the inspection robot, simultaneously breaks away from the dependence on artificial labeled data, and improves the automation level of the inspection path determination process.

Description

Technical Field

[0001] This application belongs to the field of electronic digital data processing technology, specifically relating to a method, apparatus, equipment and medium for determining inspection paths based on self-labeling learning. Background Technology

[0002] Inspecting the industrial environment is a crucial step in ensuring production safety and operational stability. With the advancement of industrial digital transformation, enterprises are gradually introducing inspection robots, leveraging their continuous operation capabilities and high-precision data acquisition to automatically monitor equipment operating status and worker behavior. In the application of inspection robots, the rationality of the inspection path directly affects inspection efficiency and the reliability of the results.

[0003] Existing inspection path planning methods primarily focus on the reachability or coverage of the inspection path during generation and execution. This makes it difficult to guarantee the continuous observability of the inspected object during the robot's movement, thus affecting the integrity and reliability of the inspection results. Although current models can automatically generate inspection paths, their optimization process still heavily relies on manually labeled data, resulting in a cumbersome and time-consuming process that reduces overall efficiency. Summary of the Invention

[0004] This application provides a method, apparatus, device, and medium for determining inspection paths based on self-labeling learning. The purpose is to ensure the continuous monitorability of the inspection sensing object during the movement of the inspection robot, while eliminating the dependence on manually labeled data and improving the automation level of the inspection path determination process.

[0005] In a first aspect, embodiments of this application provide a method for determining inspection paths based on self-labeling learning, the method comprising: The global perception information of the inspection area and the local perception information of the inspection robot are obtained, and the global perception information and the local perception information are input into the pre-built autonomous inspection decision model to obtain the initial inspection path output by the autonomous inspection decision model. The inspection robot is controlled to move along the initial inspection path and acquire inspection perception information, and the effective perception time ratio of the inspection perception objects corresponding to the initial inspection path is calculated based on the inspection perception information; wherein, the inspection perception information includes pose information and image information. If the effective perception duration ratio does not reach the preset ratio threshold, the initial inspection path is divided into multiple path segments, and the state representation parameters of the inspection perception object in each path segment are determined according to the image information in the inspection perception information. The target progressiveness of each path segment is determined according to the state representation parameters. The path segments with the target progressiveness less than the preset degree threshold are determined as deviation path segments, and the target state constraint information corresponding to the deviation path segments is constructed. The optimal inspection path is generated based on the target state constraint information corresponding to each deviation path segment, and the optimal inspection path is added to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

[0006] Furthermore, dividing the initial inspection path into multiple path segments includes: When it is detected that the effective perception state of the inspection object in the image information has changed, the initial inspection path is segmented to obtain path segments. And / or, When the pose change of the pose information is detected to reach a preset pose change threshold, the initial inspection path is segmented to obtain path segments. And / or, When the time series corresponding to the initial inspection path is identified to reach the preset segment duration, the initial inspection path is segmented to obtain path segments.

[0007] Furthermore, the state characterization parameters include perception confidence, perception area size, and perception center location; Accordingly, determining the target asymptotic degree of each path segment based on the state characterization parameters includes: For each path segment, the perception confidence, perception area size, and perception center location of the path segment are normalized to obtain standardized parameters. Based on the preset inspection targets, determine the target deviation values ​​corresponding to each standardized parameter, and calculate the weighted sum of each target deviation value to obtain the comprehensive deviation value; The difference between the comprehensive deviation value of the path segment and the comprehensive deviation value of the previous path segment is calculated and used as the target progressiveness of the path segment.

[0008] Furthermore, generating the optimal inspection path based on the target state constraint information corresponding to each deviation path segment includes: For each deviation path segment, based on the target state constraint information corresponding to the deviation path segment, and taking the initial pose information in the inspection perception information corresponding to the deviation path segment as the starting point, multiple candidate path segments are generated. Calculate the action cost index for each candidate path segment, and determine the candidate path segment with the smallest action cost index as the target path segment. Replace the deviation path segment with the target path segment in the initial inspection path to obtain the optimal inspection path.

[0009] Furthermore, the calculation of the action cost index for each candidate path segment includes: For each candidate path segment, calculate the pose change of the candidate path segment and the sensing failure of the candidate path segment; The pose change and the perception failure are normalized and weighted summed to obtain the action cost index of the candidate path segment.

[0010] Furthermore, calculating the sensing failure amount of the candidate path segment includes: Based on the pose information at each time point in the candidate path segment, predict the perception interruption interval of the inspection sensing object; The duration of the perception interruption interval is weighted and accumulated to obtain the perception failure amount of the candidate path segment.

[0011] Secondly, embodiments of this application provide an inspection path determination device based on self-labeling learning, the device comprising: The initial path determination module is used to acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into the pre-built autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model. An effective perception and recognition module is used to control the inspection robot to move along the initial inspection path and acquire inspection perception information, and to calculate the effective perception time ratio of the inspection perception objects corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information. The deviation segment determination module is used to divide the initial inspection path into multiple path segments when the effective perception duration ratio does not reach a preset ratio threshold, and determine the state representation parameters of the inspection perception object in each path segment based on the image information in the inspection perception information, determine the target progressiveness of each path segment based on the state representation parameters, determine the path segment with the target progressiveness less than a preset degree threshold as the deviation path segment, and construct the target state constraint information corresponding to the deviation path segment. The optimal path generation module is used to generate the optimal inspection path based on the target state constraint information corresponding to each deviation path segment, and add the optimal inspection path to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

[0012] Furthermore, the deviation segment determination module is specifically used for: The initial inspection path is divided into multiple path segments, and the state representation parameters of the inspection sensing object in each path segment are determined based on the image information in the inspection sensing information. The target progression level of each path segment is determined based on the state characterization parameters. The path segments where the target progressiveness is less than a preset threshold are identified as deviation path segments.

[0013] Furthermore, the deviation segment determination module is specifically used for: When it is detected that the effective perception state of the inspection object in the image information has changed, the initial inspection path is segmented to obtain path segments. And / or, When the pose change of the pose information is detected to reach a preset pose change threshold, the initial inspection path is segmented to obtain path segments. And / or, When the time series corresponding to the initial inspection path is identified to reach the preset segment duration, the initial inspection path is segmented to obtain path segments.

[0014] Furthermore, the state characterization parameters include perception confidence, perception area size, and perception center location; Accordingly, the deviation segment determination module is specifically used for: For each path segment, the perception confidence, perception area size, and perception center location of the path segment are normalized to obtain standardized parameters. Based on the preset inspection targets, determine the target deviation values ​​corresponding to each standardized parameter, and calculate the weighted sum of each target deviation value to obtain the comprehensive deviation value; The difference between the comprehensive deviation value of the path segment and the comprehensive deviation value of the previous path segment is calculated and used as the target progressiveness of the path segment.

[0015] Furthermore, the optimal path generation module is specifically used for: For each deviation path segment, based on the target state constraint information corresponding to the deviation path segment, and taking the initial pose information in the inspection perception information corresponding to the deviation path segment as the starting point, multiple candidate path segments are generated. Calculate the action cost index for each candidate path segment, and determine the candidate path segment with the smallest action cost index as the target path segment. Replace the deviation path segment with the target path segment in the initial inspection path to obtain the optimal inspection path.

[0016] Furthermore, the optimal path generation module is specifically used for: For each candidate path segment, calculate the pose change of the candidate path segment and the sensing failure of the candidate path segment; The pose change and the perception failure are normalized and weighted summed to obtain the action cost index of the candidate path segment.

[0017] Furthermore, the optimal path generation module is specifically used for: Based on the pose information at each time point in the candidate path segment, predict the perception interruption interval of the inspection sensing object; The duration of the perception interruption interval is weighted and accumulated to obtain the perception failure amount of the candidate path segment.

[0018] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method described in the first aspect.

[0019] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the method described in the first aspect.

[0020] In this embodiment, global perception information of the inspection area and local perception information of the inspection robot are acquired, and the global perception information and the local perception information are input into a pre-constructed autonomous inspection decision-making model to obtain an initial inspection path output by the autonomous inspection decision-making model; the inspection robot is controlled to move according to the initial inspection path and acquire inspection perception information, and the effective perception time ratio of the inspection perception object corresponding to the initial inspection path is calculated based on the inspection perception information; wherein, the inspection perception information includes pose information and image information; if the effective perception time ratio does not reach a preset ratio threshold, The initial inspection path is divided into multiple path segments. Based on the image information in the inspection perception information, the state representation parameters of the inspection perception object in each path segment are determined. The target progressiveness of each path segment is determined based on the state representation parameters. Path segments with a target progressiveness less than a preset threshold are identified as deviation path segments, and target state constraint information corresponding to these deviation path segments is constructed. An optimal inspection path is generated based on the target state constraint information corresponding to each deviation path segment, and this optimal inspection path is added to the training dataset of the inspection autonomous decision-making model for training and updating the model. This self-labeled learning-based inspection path determination method ensures the continuous monitorability of the inspection perception object during the inspection robot's movement, while eliminating reliance on manually labeled data and improving the automation level of the inspection path determination process. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating an inspection path determination method based on self-labeling learning provided in an embodiment of this application. Figure 2 This is a flowchart illustrating another inspection path determination method based on self-labeling learning provided in an embodiment of this application. Figure 3 This is a flowchart illustrating another inspection path determination method based on self-labeling learning provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of an inspection path determination device based on self-labeling learning provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0023] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0024] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0025] The following description, in conjunction with the accompanying drawings, details the inspection path determination method, apparatus, equipment, and medium based on self-labeling learning provided in this application through specific embodiments and application scenarios.

[0026] First, this application applies to scenarios requiring autonomous control of inspection robots to complete perception and monitoring, such as equipment inspection in industrial settings, urban infrastructure inspection, and warehouse and logistics environment inspection. Based on the above application scenarios, it is understood that the executing entity of this application can be an intelligent terminal device with image processing capabilities, such as an industrial-grade edge computing terminal, an intelligent monitoring host, a portable industrial control computer, a robot-specific controller, and an industrial IoT edge gateway, etc., without further limitations.

[0027] Figure 1 This is a flowchart illustrating a self-labeling learning-based inspection path determination method provided in an embodiment of this application. Figure 1 As shown, the specific steps include the following: S101, acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into the pre-built autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model.

[0028] The inspection area can be the spatial range in which the inspection robot needs to perform autonomous perception and monitoring tasks; the global perception information can be image information covering the entire inspection area.

[0029] Among them, the inspection robot can be an intelligent robot with autonomous mobility and equipped with multiple sensors, used to replace or assist humans in completing inspection tasks, such as an embodied intelligent robot; the local perception information can be information reflecting the real-time situation around the inspection robot, which can include the robot's pose information and the image information collected by the inspection robot.

[0030] In one embodiment, the method for acquiring global perception information of the inspection area and local perception information of the inspection robot can be achieved by using a multi-camera network monitoring system to acquire global perception information; simultaneously, local perception information can be collected in real time by the positioning module and image sensor mounted on the inspection robot itself. The multi-camera network monitoring system can be a visual monitoring system deployed at key locations within the inspection area, achieving full coverage and blind-spot-free data collection through the collaborative work of multiple cameras.

[0031] Among them, the autonomous decision-making model for inspection can be an intelligent algorithm model that can autonomously output path planning results that meet the requirements of the inspection task based on the input global perception information and local perception information, such as the VLA (Vision-Language-Action) large model. Specifically, the VLA large model can be a Transformer architecture network that integrates a visual encoder, a language encoder, and an action decoder, and can receive multimodal input and directly output path point sequences or control commands.

[0032] In one embodiment, the method of pre-building the autonomous decision-making model for inspection can be to build a model network structure suitable for inspection path planning, collect a large amount of global perception information, local perception information and corresponding expert trajectory data of the inspection scenario, and pre-train it using a behavior cloning supervised learning algorithm; or it can directly use a pre-trained VLA large model.

[0033] The inspection path can be a desired trajectory consisting of a series of continuous pose points and the timestamps corresponding to each pose point. Correspondingly, the initial inspection path is the inspection path initially output by the autonomous inspection decision-making model and has not been optimized or verified.

[0034] Understandably, the input data for the autonomous decision-making model of the inspection can also include task intent description information and prior environmental information of the inspection area.

[0035] In one embodiment, the method of inputting global perception information and local perception information into a pre-built autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model can be achieved by encoding global perception information into tensor feature maps, encoding local perception information, task intent description information and environmental prior information into feature vectors respectively, performing cross-modal attention fusion within the autonomous inspection decision-making model, and finally generating the initial inspection path by autoregression of the action decoder.

[0036] S102, control the inspection robot to move along the initial inspection path and acquire inspection perception information, and calculate the effective perception time ratio of the inspection perception object corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information.

[0037] The inspection perception information can be the real-time state and perception data recorded by the inspection robot as it moves along the initial inspection path. This information can include pose information and image information. Specifically, pose information can be the three-dimensional coordinates and attitude angles of the inspection robot at each sampling moment; image information can be the visual image data collected by the image sensor on the inspection robot at the corresponding sampling moment.

[0038] In one embodiment, the method of controlling the inspection robot to move along the initial inspection path and acquire inspection perception information can be achieved by sending the initial inspection path to the underlying motion controller of the inspection robot, driving the inspection robot to track the initial inspection path, and recording the real-time pose of the inspection robot and the images captured by the camera at a fixed frequency.

[0039] Among them, the inspection sensing object corresponding to the initial inspection path can be the specific target entity that needs to be monitored in this inspection task, which is determined by the inspection autonomous decision-making model. The effective sensing time ratio can be the ratio of the cumulative time during which the inspection sensing object appears clearly and unobstructed in the image information to the total movement time during the entire path movement.

[0040] In one embodiment, the effective perception time ratio of the inspection perception object corresponding to the initial inspection path can be obtained by running a target detection algorithm on each frame of image information collected, judging and counting the proportion of the number of frames in which the inspection perception object is located in the image information to the total number of frames, and thus obtaining the effective perception time ratio.

[0041] S103, if the effective perception duration ratio does not reach the preset ratio threshold, the initial inspection path is divided into multiple path segments, and the state representation parameters of the inspection perception object in each path segment are determined according to the image information in the inspection perception information. The target progressiveness of each path segment is determined according to the state representation parameters. The path segment with the target progressiveness less than the preset degree threshold is determined as the deviation path segment, and the target state constraint information corresponding to the deviation path segment is constructed.

[0042] The preset ratio threshold can be set according to the accuracy requirements of the inspection task, and is a critical value for judging whether the continuous perception effect of the initial inspection path is qualified, such as 0.75. If the effective perception time ratio does not reach the preset ratio threshold, it means that the inspection robot's perception coverage of the inspection object is insufficient during the movement, so the initial inspection path needs to be optimized to enhance the perception continuity.

[0043] In one embodiment, the method of dividing the initial inspection path into multiple path segments can be to divide the initial inspection path into a preset number of path segments on an average basis according to the path length.

[0044] Among them, the state characterization parameters can be indicators used to quantify the perception quality of the inspected sensing object in the image information.

[0045] In one embodiment, the method of determining the state representation parameters of the inspection and sensing object in each path segment based on the image information in the inspection and sensing information can be to run a target detection algorithm on the image information in the inspection and sensing information and determine the ratio of the number of pixels representing the inspection and sensing object in the image information to the total number of pixels in the image information as the state representation parameters of the path segment.

[0046] The target asymptotic rate can be the rate at which the state representation parameters of the path segment progress towards the preset inspection target. Specifically, the preset inspection target can be the state representation parameters of the inspection object in the image information, output by the inspection autonomous decision-making model, when the ideal inspection robot reaches the end of the initial inspection path, that is, when it reaches the final perception pose.

[0047] In one embodiment, the method for determining the target progressiveness of each path segment based on the state characterization parameters can be as follows: subtract the state characterization parameters corresponding to the starting point of the path segment from the state characterization parameters corresponding to the end point of the path segment to obtain the change in the state characterization parameters of the path segment; divide the change in the state characterization parameters by the preset inspection target to obtain the target progressiveness of the path segment.

[0048] The state characterization parameters may include perception confidence, perception area size, and perception center location.

[0049] Accordingly, in one embodiment, determining the target progressiveness of each path segment based on the state characterization parameters includes: for each path segment, normalizing the perception confidence, perception area size, and perception center position of the path segment to obtain standardized parameters; determining the target deviation value corresponding to each standardized parameter based on a preset inspection target, and performing a weighted summation of each target deviation value to obtain a comprehensive deviation value; and calculating the difference between the comprehensive deviation value of the path segment and the comprehensive deviation value of the previous path segment as the target progressiveness of the path segment.

[0050] Among them, the perception confidence can be a quantitative indicator that reflects the degree of identifiability of the inspection and perception object in the image information; the perception area size can be the size of the pixel area occupied by the inspection and perception object in the image information; and the perception center position can be the coordinates of the center of the pixel area occupied by the inspection and perception object in the image information.

[0051] In one embodiment, the method for determining the perception confidence, perception area size, and perception center location of a path segment can be as follows: run a target detection algorithm on the image information of the end point of the path segment, extract the pixel area occupied by the inspection and perception object, calculate the image clarity of the occupied pixel area as the perception confidence, determine the rectangular bounding box of the occupied pixel area, calculate the size of the rectangular bounding box as the perception area size, and determine the coordinate center of the rectangular bounding box as the perception center location.

[0052] The standardized parameters can be dimensionless values ​​obtained by normalizing the original perception confidence, perception area size, and perception center location.

[0053] In one embodiment, the normalization of the perception confidence, perception region size, and perception center position of the path segment to obtain the normalized parameters can be achieved by calculating the ratio of the image sharpness of the occupied pixel region to the image sharpness of the entire image information as the normalized parameter corresponding to the perception confidence; calculating the ratio of the perception region size to the size of the entire image information as the normalized parameter corresponding to the perception region size; calculating the distance between the perception center position and the image information center position; calculating the longest distance between the image information edge and the image information center position; and calculating the ratio of this distance to the longest distance as the normalized parameter corresponding to the perception center position.

[0054] The preset inspection targets may include the ideal normalized perception confidence level, the ideal normalized perception area size, and the ideal normalized perception center location.

[0055] Correspondingly, the target deviation value can be the absolute difference between each standardized parameter and the corresponding ideal value in the preset inspection target; the comprehensive deviation value can be a single value obtained by weighting and summing multiple target deviation values ​​according to preset weights.

[0056] In one embodiment, the method of determining the target deviation value corresponding to each standardized parameter according to the preset inspection target and calculating the weighted sum of each target deviation value to obtain the comprehensive deviation value can be achieved by calculating the absolute difference between each standardized parameter and the corresponding ideal value in the preset inspection target, multiplying each absolute difference by the corresponding preset weight, and finally summing the multiplication results to obtain the comprehensive deviation value.

[0057] In one embodiment, the method of calculating the difference between the comprehensive deviation value of a path segment and the comprehensive deviation value of the previous path segment as the target progressiveness of the path segment is to calculate the difference between the comprehensive deviation value corresponding to the image information at the end of the path segment and the comprehensive deviation value corresponding to the image information at the end of the previous path segment as the target progressiveness of the path segment.

[0058] The advantage of this approach is that by calculating a comprehensive deviation value based on perception confidence, perception area size, and perception center location, multi-dimensional perception performance indicators can be integrated into a single quantifiable degree of deviation. This allows for the accurate identification of deviation path segments where perception performance improvement is slow, providing an objective basis for subsequent path optimization and model self-labeling learning.

[0059] The preset threshold can be the minimum acceptable level of progress set in advance.

[0060] Among them, the deviation path segment can be a continuous sub-path in the initial inspection path where the pose combination of the inspection robot relative to the inspection and sensing object leads to poor continuous sensing effect.

[0061] If the target progression of a path segment is less than a preset threshold, it means that the target progression is too low. In other words, the speed at which the state representation parameters of the inspected and sensed object in the path segment approach the ideal target is insufficient, and the sensing effect is not improved with the expected efficiency. Therefore, the path segment can be identified as a deviation path segment.

[0062] The target state constraint information can be a set of constraints used to guide the optimization of deviation path segments, including effective perception duration ratio constraints and end-point pose continuity constraints. Specifically, the effective perception duration ratio constraint means that the effective perception duration ratio of the inspected sensing object must reach a preset ratio threshold; the end-point pose continuity constraint means that the pose information at the end of the deviation path segment must be consistent with the original or consistent with the pose information at the beginning of the adjacent path segment.

[0063] In one embodiment, the method for constructing the target state constraint information corresponding to the deviation path segment can be to determine the target state constraint information by using a preset proportional threshold and the current endpoint pose of the deviation path segment.

[0064] S104, Generate the optimal inspection path according to the target state constraint information corresponding to each deviation path segment, and add the optimal inspection path to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

[0065] The optimal inspection path can be an inspection path that satisfies the target state constraint information corresponding to all deviation path segments, while achieving the best continuous perception effect on the inspection object.

[0066] In one embodiment, the method of generating the optimal inspection path based on the target state constraint information corresponding to each deviation path segment can be as follows: for each deviation path segment, take its starting pose as the initial state, and take the effective perception time ratio and the ending pose as constraints. Use a local path planning algorithm to generate an optimized path segment that satisfies all constraints. Then, all optimized path segments and the path segments that do not need optimization in the initial inspection path are sequentially spliced ​​together to obtain the complete optimal inspection path.

[0067] The training dataset for the autonomous inspection decision-making model can be an incremental database containing a large number of input-output sample pairs. The input of each sample pair includes global perception information of the inspection area and local perception information of the inspection robot, and the output corresponds to the optimal inspection path.

[0068] In one embodiment, the optimal inspection path is added to the training dataset of the autonomous inspection decision-making model for training and updating the autonomous inspection decision-making model. This can be done by encapsulating the current global perception information, local perception information, and the optimal inspection path into a training sample and storing it in the cache pool of the training dataset. When the number of newly added training samples in the cache pool reaches a preset threshold or after a fixed time interval, incremental learning or experience replay methods are used to fine-tune the autonomous inspection decision-making model.

[0069] In this embodiment, global perception information of the inspection area and local perception information of the inspection robot are acquired, and the global perception information and the local perception information are input into a pre-constructed autonomous inspection decision-making model to obtain an initial inspection path output by the autonomous inspection decision-making model; the inspection robot is controlled to move according to the initial inspection path and acquire inspection perception information, and the effective perception time ratio of the inspection perception object corresponding to the initial inspection path is calculated based on the inspection perception information; wherein, the inspection perception information includes pose information and image information; if the effective perception time ratio does not reach a preset ratio threshold, The initial inspection path is divided into multiple path segments. Based on the image information in the inspection perception information, the state representation parameters of the inspection perception object in each path segment are determined. The target progressiveness of each path segment is determined based on the state representation parameters. Path segments with a target progressiveness less than a preset threshold are identified as deviation path segments, and target state constraint information corresponding to these deviation path segments is constructed. An optimal inspection path is generated based on the target state constraint information corresponding to each deviation path segment, and this optimal inspection path is added to the training dataset of the inspection autonomous decision-making model for training and updating the model. This self-labeled learning-based inspection path determination method ensures the continuous monitorability of the inspection perception object during the inspection robot's movement, while eliminating reliance on manually labeled data and improving the automation level of the inspection path determination process.

[0070] Figure 2 This is a flowchart illustrating another inspection path determination method based on self-labeling learning provided in an embodiment of this application. For example... Figure 2 As shown, the specific steps include the following: S201, acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into the pre-built autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model.

[0071] S202, control the inspection robot to move along the initial inspection path and acquire inspection perception information, and calculate the effective perception time ratio of the inspection perception object corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information.

[0072] S203, when the effective perception duration ratio does not reach a preset ratio threshold, and when it is detected that the effective perception state of the inspection object in the image information has changed, the initial inspection path is segmented to obtain a path segment; and / or, when it is detected that the pose change amount of the pose information reaches a preset pose change threshold, the initial inspection path is segmented to obtain a path segment; and / or, when it is detected that the time series corresponding to the initial inspection path reaches a preset segmentation duration, the initial inspection path is segmented to obtain a path segment.

[0073] Among them, the effective perception state can be whether the inspection and sensing object appears in the image information; the effective perception state of the inspection and sensing object in the image information changes, which can be the appearance of the inspection and sensing object in the image information or the disappearance of the inspection and sensing object.

[0074] When the effective perception state of the inspection object in the image information changes, it indicates that the relative pose relationship between the inspection robot and the inspection object has changed, resulting in a change that makes the perception state effective or ineffective. Therefore, the initial inspection path can be segmented to obtain path segments.

[0075] Among them, the pose change can be the comprehensive change after weighted combination of the position change and the attitude angle change of the inspection robot; the preset pose change threshold can be the maximum allowable pose change set according to the task requirements.

[0076] When the pose change of the pose information reaches the preset pose change threshold, it means that the inspection robot has moved a sufficiently large distance or rotated a sufficiently large angle, causing the environmental perception context to change significantly. Therefore, the initial inspection path can be segmented to obtain path segments.

[0077] The time series can be a sequence of timestamps corresponding to each pose point on the initial inspection path; the preset segment duration can be a fixed time interval length set in advance, such as 2 seconds, 5 seconds or 10 seconds.

[0078] The time series corresponding to the initial inspection path reaches the preset segment duration, indicating that the inspection robot has been moving continuously for a fixed period of time. In order to maintain the uniformity of time granularity of each path segment and facilitate subsequent statistics and analysis, the initial inspection path can be segmented to obtain path segments.

[0079] S204, determine the state representation parameters of the inspection and sensing object in each path segment based on the image information in the inspection and sensing information, determine the target progressiveness of each path segment based on the state representation parameters, determine the path segment with the target progressiveness less than a preset threshold as the deviation path segment, and construct the target state constraint information corresponding to the deviation path segment.

[0080] S205, generate the optimal inspection path according to the target state constraint information corresponding to each deviation path segment, and add the optimal inspection path to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

[0081] The advantage of this scheme is that by using a variety of complementary segmentation strategies based on changes in perception state, changes in pose, and fixed time intervals, it can flexibly adapt to different inspection scenarios and perception task requirements. It can perform fine segmentation at key locations where perception effectiveness changes abruptly, and can also ensure the controllability and consistency of segmentation granularity through pose changes and time-uniform segmentation.

[0082] Figure 3 This is a flowchart illustrating another inspection path determination method based on self-labeling learning provided in this application. Figure 3 As shown, the specific steps include the following: S301, acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into the pre-built autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model.

[0083] S302, control the inspection robot to move along the initial inspection path and acquire inspection perception information, and calculate the effective perception time ratio of the inspection perception object corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information.

[0084] S303, if the effective perception duration ratio does not reach a preset ratio threshold, the initial inspection path is divided into multiple path segments, and the state representation parameters of the inspection perception object in each path segment are determined according to the image information in the inspection perception information. The target progressiveness of each path segment is determined according to the state representation parameters. The path segments with the target progressiveness less than a preset degree threshold are determined as deviation path segments, and the target state constraint information corresponding to the deviation path segments is constructed.

[0085] S304, For each deviation path segment, based on the target state constraint information corresponding to the deviation path segment, and taking the initial pose information in the inspection perception information corresponding to the deviation path segment as the starting point, multiple candidate path segments are generated.

[0086] Among them, the candidate path segment can be multiple alternative sub-paths starting from the starting point of the deviation path segment.

[0087] In one embodiment, based on the target state constraint information corresponding to the deviation path segment, and starting from the initial pose information in the inspection and perception information corresponding to the deviation path segment, multiple candidate path segments are generated. This can be achieved by using the initial pose information in the inspection and perception information corresponding to the deviation path segment as the initial state, and using the effective perception time ratio and the endpoint pose as constraints, and using a local path planning algorithm to generate path segments that satisfy all constraints as candidate path segments.

[0088] S305, calculate the action cost index of each candidate path segment, and determine the candidate path segment with the smallest action cost index as the target path segment. Replace the deviation path segment with the target path segment in the initial inspection path to obtain the optimal inspection path.

[0089] Among them, the action cost index can be a quantitative value used to evaluate the quality of candidate path segments.

[0090] In one embodiment, the action cost index for each candidate path segment can be calculated by using the path length of each candidate path segment as the action cost index.

[0091] In one embodiment, calculating the action cost index for each candidate path segment includes: calculating the pose change of the candidate path segment and the perception failure of the candidate path segment for each candidate path segment; normalizing the pose change and the perception failure respectively and performing weighted summation to obtain the action cost index of the candidate path segment.

[0092] Among them, the change in pose can be the combined change of the position change and the change in attitude angle of the inspection robot.

[0093] In one embodiment, the pose change of a candidate path segment can be calculated by using the path length of the candidate path segment as the position change, the absolute value of the yaw angle change as the attitude angle change, and then weighting and summing the position change and attitude angle change according to a preset weight to obtain the comprehensive change.

[0094] Among them, the sensing failure quantity can be the duration of time that the inspected sensing object is in an invalid sensing state in the candidate path segment.

[0095] In one embodiment, the method for calculating the sensing failure amount of a candidate path segment can be to identify whether the object being inspected falls within the field of view of the image sensor at each moment based on the pose information of the candidate path segment at each moment and the prior environmental information of the inspection area, and to accumulate the duration during which the object being inspected does not fall within the field of view of the image sensor as the sensing failure amount of the candidate path segment.

[0096] In one embodiment, calculating the perceived failure amount of the candidate path segment includes: Based on the pose information at each time point in the candidate path segment, predict the perception interruption interval of the inspection sensing object; The duration of the perception interruption interval is weighted and accumulated to obtain the perception failure amount of the candidate path segment.

[0097] The perception interruption interval can be the time period during which the inspected sensing object is continuously in an invalid sensing state during the movement of the candidate path segment.

[0098] In one embodiment, the method of predicting the perception interruption interval of the inspection and sensing object based on the pose information at each moment in the candidate path segment can be achieved by identifying whether the inspection and sensing object falls within the field of view of the image sensor at each moment based on the pose information at each moment in the candidate path segment and the environmental prior information of the inspection area. The moment when the continuous inspection and sensing object does not fall within the field of view of the image sensor is determined as the initial interruption interval, and the initial interruption interval with a duration exceeding a preset duration is determined as the perception interruption interval.

[0099] In one embodiment, the method of obtaining the sensing failure amount of a candidate path segment by weighted summation of the durations of the sensing interruption intervals can be achieved by multiplying the duration of each sensing interruption interval by its corresponding preset weight, and then summing the results. The preset weights of each sensing interruption interval are set in ascending order according to their temporal sequence within the candidate path segment; that is, the closer a sensing interruption interval is to the end of the candidate path segment, the larger its corresponding preset weight. For example, the preset weight of each sensing interruption interval is equal to the sum of the preset weight of the previous sensing interruption interval and a preset ascending weight variable.

[0100] The advantage of this scheme is that by using a weighted accumulation of perception interruption duration over time, it can highlight the greater impact of later perception interruptions on the perception effect, making the perception failure rate more in line with the actual inspection scenario requirements.

[0101] In one embodiment, the pose change and perception failure are normalized and weighted summed to obtain the action cost index of the candidate path segment. This can be achieved by statistically analyzing the maximum and minimum values ​​of pose change in all candidate path segments, performing minimum-maximum normalization on the pose change of each candidate path segment, and similarly performing minimum-maximum normalization on the perception failure. Finally, the pose change and perception failure are multiplied by their respective preset weights, and the results are summed to obtain the action cost index of the candidate path segment.

[0102] The advantage of this scheme is that the action cost index no longer only focuses on the movement efficiency of the path, but can also directly reflect the path's ability to ensure continuous perception. Thus, when selecting the best candidate path segment, it can automatically favor those paths that, even if they are slightly longer, can maintain clear perception.

[0103] The target path segment can be the optimal candidate path segment selected from all candidate path segments.

[0104] Among them, the candidate path segment with the lowest action cost index indicates that the candidate path segment has the best overall performance under the premise of satisfying the target state constraint information. Therefore, the candidate path segment can be determined as the target path segment.

[0105] In one embodiment, the optimal inspection path is obtained by replacing the deviation path segment with the target path segment in the initial inspection path. This can be achieved by deleting the original pose point sequence from the start point to the end point of the deviation path segment in the initial inspection path and inserting the pose point sequence of the target path segment.

[0106] S306, The optimal inspection path is added to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

[0107] The advantage of this approach is that by generating multiple candidate path segments and automatically selecting the optimal target path segment based on the action cost index, the path with the lowest overall cost can be automatically selected from numerous candidate solutions in a quantitative manner, avoiding the inefficiency and uncertainty caused by manual trial and error or random selection.

[0108] Figure 4 This is a schematic diagram of a self-labeling learning-based inspection path determination device provided in an embodiment of this application. Figure 4 As shown, the device includes: The initial path determination module 410 is used to acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into a pre-built autonomous inspection decision model to obtain the initial inspection path output by the autonomous inspection decision model. The effective perception and recognition module 420 is used to control the inspection robot to move along the initial inspection path and acquire inspection perception information, and to calculate the effective perception time ratio of the inspection perception objects corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information. The deviation segment determination module 430 is used to divide the initial inspection path into multiple path segments when the effective perception duration ratio does not reach a preset ratio threshold, and determine the state representation parameters of the inspection perception object in each path segment based on the image information in the inspection perception information, determine the target progressiveness of each path segment based on the state representation parameters, determine the path segment with the target progressiveness less than a preset degree threshold as the deviation path segment, and construct the target state constraint information corresponding to the deviation path segment. The optimal path generation module 440 is used to generate the optimal inspection path according to the target state constraint information corresponding to each deviation path segment, and add the optimal inspection path to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

[0109] Furthermore, the deviation segment determination module 430 is specifically used for: When it is detected that the effective perception state of the inspection object in the image information has changed, the initial inspection path is segmented to obtain path segments. And / or, When the pose change of the pose information is detected to reach a preset pose change threshold, the initial inspection path is segmented to obtain path segments. And / or, When the time series corresponding to the initial inspection path is identified to reach the preset segment duration, the initial inspection path is segmented to obtain path segments.

[0110] Furthermore, the state characterization parameters include perception confidence, perception area size, and perception center location; Accordingly, the deviation segment determination module 430 is specifically used for: For each path segment, the perception confidence, perception area size, and perception center location of the path segment are normalized to obtain standardized parameters. Based on the preset inspection targets, determine the target deviation values ​​corresponding to each standardized parameter, and calculate the weighted sum of each target deviation value to obtain the comprehensive deviation value; The difference between the comprehensive deviation value of the path segment and the comprehensive deviation value of the previous path segment is calculated and used as the target progressiveness of the path segment.

[0111] Furthermore, the optimal path generation module 440 is specifically used for: For each deviation path segment, based on the target state constraint information corresponding to the deviation path segment, and taking the initial pose information in the inspection perception information corresponding to the deviation path segment as the starting point, multiple candidate path segments are generated. Calculate the action cost index for each candidate path segment, and determine the candidate path segment with the smallest action cost index as the target path segment. Replace the deviation path segment with the target path segment in the initial inspection path to obtain the optimal inspection path.

[0112] Furthermore, the optimal path generation module 440 is specifically used for: For each candidate path segment, calculate the pose change of the candidate path segment and the sensing failure of the candidate path segment; The pose change and the perception failure are normalized and weighted summed to obtain the action cost index of the candidate path segment.

[0113] Furthermore, the optimal path generation module 440 is specifically used for: Based on the pose information at each time point in the candidate path segment, predict the perception interruption interval of the inspection sensing object; The duration of the perception interruption interval is weighted and accumulated to obtain the perception failure amount of the candidate path segment.

[0114] In this embodiment, the initial path determination module is used to acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into a pre-constructed autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model; the effective perception recognition module is used to control the inspection robot to move according to the initial inspection path and acquire inspection perception information, and to calculate the effective perception time ratio of the inspection perception objects corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information; the deviation segment determination module is used to determine the deviation segment when the effective perception time ratio does not reach the predetermined value. With a set percentage threshold, the initial inspection path is divided into multiple path segments. Based on the image information in the inspection perception information, the state representation parameters of the inspection perception object in each path segment are determined. The target progressiveness of each path segment is determined based on the state representation parameters. Path segments with a target progressiveness less than a preset threshold are identified as deviation path segments, and target state constraint information corresponding to these deviation path segments is constructed. An optimal path generation module generates an optimal inspection path based on the target state constraint information corresponding to each deviation path segment and adds the optimal inspection path to the training dataset of the inspection autonomous decision-making model for training and updating the model. This self-labeled learning-based inspection path determination device ensures the continuous monitorability of the inspection perception object during the inspection robot's movement, while eliminating reliance on manually labeled data and improving the automation level of the inspection path determination process.

[0115] The inspection path determination device based on self-labeling learning in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0116] The inspection path determination device based on self-labeling learning in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0117] The inspection path determination device based on self-labeling learning provided in this application can realize the various processes implemented in the above embodiments. To avoid repetition, it will not be described again here.

[0118] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example... Figure 5 As shown, this application embodiment also provides an electronic device 500, including a processor 501, a memory 502, and a program or instructions stored in the memory 502 and executable on the processor 501. When the program or instructions are executed by the processor 501, they implement the various processes of the above-described self-labeling learning-based inspection path determination embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0119] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0120] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described self-labeling learning-based inspection path determination embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0121] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0122] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0124] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0125] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A method for determining inspection paths based on self-labeling learning, characterized in that, The method includes: The global perception information of the inspection area and the local perception information of the inspection robot are obtained, and the global perception information and the local perception information are input into the pre-built autonomous inspection decision model to obtain the initial inspection path output by the autonomous inspection decision model. The inspection robot is controlled to move along the initial inspection path and acquire inspection perception information, and the effective perception time ratio of the inspection perception objects corresponding to the initial inspection path is calculated based on the inspection perception information; wherein, the inspection perception information includes pose information and image information. If the effective perception duration ratio does not reach the preset ratio threshold, the initial inspection path is divided into multiple path segments, and the state representation parameters of the inspection perception object in each path segment are determined according to the image information in the inspection perception information. The target progressiveness of each path segment is determined according to the state representation parameters. The path segments with the target progressiveness less than the preset degree threshold are determined as deviation path segments, and the target state constraint information corresponding to the deviation path segments is constructed. The optimal inspection path is generated based on the target state constraint information corresponding to each deviation path segment, and the optimal inspection path is added to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

2. The inspection path determination method based on self-labeling learning according to claim 1, characterized in that, The process of dividing the initial inspection path into multiple path segments includes: When it is detected that the effective perception state of the inspection object in the image information has changed, the initial inspection path is segmented to obtain path segments. And / or, When the pose change of the pose information is detected to reach a preset pose change threshold, the initial inspection path is segmented to obtain path segments. And / or, When the time series corresponding to the initial inspection path is identified to reach the preset segment duration, the initial inspection path is segmented to obtain path segments.

3. The inspection path determination method based on self-labeling learning according to claim 1, characterized in that, The state characterization parameters include perception confidence, perception area size, and perception center location. Accordingly, determining the target asymptotic degree of each path segment based on the state characterization parameters includes: For each path segment, the perception confidence, perception area size, and perception center location of the path segment are normalized to obtain standardized parameters. Based on the preset inspection targets, determine the target deviation values ​​corresponding to each standardized parameter, and calculate the weighted sum of each target deviation value to obtain the comprehensive deviation value; The difference between the comprehensive deviation value of the path segment and the comprehensive deviation value of the previous path segment is calculated and used as the target progressiveness of the path segment.

4. The inspection path determination method based on self-labeling learning according to claim 1, characterized in that, The step of generating the optimal inspection path based on the target state constraint information corresponding to each deviation path segment includes: For each deviation path segment, based on the target state constraint information corresponding to the deviation path segment, and taking the initial pose information in the inspection perception information corresponding to the deviation path segment as the starting point, multiple candidate path segments are generated. Calculate the action cost index for each candidate path segment, and determine the candidate path segment with the smallest action cost index as the target path segment. Replace the deviation path segment with the target path segment in the initial inspection path to obtain the optimal inspection path.

5. The inspection path determination method based on self-labeling learning according to claim 4, characterized in that, The calculation of the action cost index for each candidate path segment includes: For each candidate path segment, calculate the pose change of the candidate path segment and the sensing failure of the candidate path segment; The pose change and the perception failure are normalized and weighted summed to obtain the action cost index of the candidate path segment.

6. The inspection path determination method based on self-labeling learning according to claim 5, characterized in that, The calculation of the sensing failure amount of the candidate path segment includes: Based on the pose information at each time point in the candidate path segment, predict the perception interruption interval of the inspection sensing object; The duration of the perception interruption interval is weighted and accumulated to obtain the perception failure amount of the candidate path segment.

7. A device for determining inspection paths based on self-labeling learning, characterized in that, The device includes: The initial path determination module is used to acquire global perception information of the inspection area and local perception information of the inspection robot, and input the global perception information and the local perception information into the pre-built autonomous inspection decision-making model to obtain the initial inspection path output by the autonomous inspection decision-making model. An effective perception and recognition module is used to control the inspection robot to move along the initial inspection path and acquire inspection perception information, and to calculate the effective perception time ratio of the inspection perception objects corresponding to the initial inspection path based on the inspection perception information; wherein, the inspection perception information includes pose information and image information. The deviation segment determination module is used to divide the initial inspection path into multiple path segments when the effective perception duration ratio does not reach a preset ratio threshold, and determine the state representation parameters of the inspection perception object in each path segment based on the image information in the inspection perception information, determine the target progressiveness of each path segment based on the state representation parameters, determine the path segment with the target progressiveness less than a preset degree threshold as the deviation path segment, and construct the target state constraint information corresponding to the deviation path segment. The optimal path generation module is used to generate the optimal inspection path based on the target state constraint information corresponding to each deviation path segment, and add the optimal inspection path to the training dataset of the inspection autonomous decision-making model for training and updating the inspection autonomous decision-making model.

8. The inspection path determination device based on self-labeling learning according to claim 7, characterized in that, The deviation section determination module is specifically used for: When it is detected that the effective perception state of the inspection object in the image information has changed, the initial inspection path is segmented to obtain path segments. And / or, When the pose change of the pose information is detected to reach a preset pose change threshold, the initial inspection path is segmented to obtain path segments. And / or, When the time series corresponding to the initial inspection path is identified to reach the preset segment duration, the initial inspection path is segmented to obtain path segments.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the inspection path determination method based on self-labeling learning as described in any one of claims 1-6.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a processor, implement the inspection path determination method based on self-labeling learning as described in any one of claims 1-6.

Images