Mineral separation workshop inspection method based on unmanned aerial vehicle

Abstract

The invention discloses a mineral separation workshop inspection method based on an unmanned aerial vehicle, and relates to the technical field of mineral separation workshop inspection, and the method specifically comprises the following steps: when an image collected by the unmanned aerial vehicle is affected by oblique light irradiation, recognizing a region with shadow interference characteristics in the image, and uniformly dividing the region into a plurality of sub-regions; extracting image disturbance feature information of each sub-region from the image acquired by the unmanned aerial vehicle, analyzing the image disturbance feature information, and determining a shadow gray scale disturbance degree and a disturbance type of each sub-region; and according to the disturbance type of each sub-region, respectively executing a trend judgment strategy of the corresponding crack identification path, and carrying out dynamic regulation and control. According to the method, the problems of crack and shadow confusion and path judgment deviation caused by oblique light irradiation when the unmanned aerial vehicle obliquely shoots the preparation equipment are solved, classification identification and dynamic path regulation and control based on image disturbance intensity are realized, and the accuracy of crack identification, the stability of path judgment and the intelligent degree of inspection tasks are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of inspection tours in ore dressing workshops, and particularly relates to a method for inspecting ore dressing workshops based on unmanned aerial vehicles (UAVs). Background Art

[0002] The ore dressing workshop is an important link in mine production, responsible for separating the required minerals from the raw ore through a series of treatment processes such as crushing, screening, flotation, etc. In the ore dressing workshop, the operating status of equipment and facilities directly affects production efficiency and safety. Therefore, regularly inspecting the ore dressing workshop to ensure the normal operation of equipment and promptly discover potential failures or safety hazards is crucial for ensuring the stability and safety of the production process. The inspection tasks include comprehensively monitoring and detecting facilities, equipment, pipelines, etc. The UAV combined with fixed cameras, sensors, and AI intelligent analysis technology can achieve all-round intelligent inspection of the workshop. Through the automatic flight and real-time data collection of the UAV, temperature monitoring and status analysis can be carried out on key equipment, and the data can be transmitted to the control platform in real time. The intelligent system of the UAV can automatically identify abnormal situations and judge the operating status and potential safety hazards of the equipment through AI analysis, thus providing decision-making support for equipment maintenance and production safety. By achieving the goals of unmanned, intelligent, and automated inspection tours, the UAV technology significantly improves the efficiency and accuracy of inspection tours and can cover areas that cannot be reached by traditional manual inspection tours, ensuring the continuous and safe operation of the ore dressing workshop.

[0003] The existing drone-based inspection technology for ore dressing workshops combines drone automatic flight, sensors, cameras, and AI intelligent analysis systems to achieve intelligent inspection of all links in the ore dressing workshop. First, the drone automatically flies along a preset route, covering all areas of the ore dressing workshop to ensure no omission. High-precision sensors and infrared temperature detectors carried on the drone continuously monitor the operating status of equipment in the workshop, such as key parameters like temperature, humidity, and pressure of pipelines, tanks, motors, ball mills, etc. At the same time, fixed cameras and the drone's camera system obtain real-time video image data of the equipment and the environment. Combining with AI image recognition technology, it identifies whether there are abnormalities in the equipment, such as electric leakage, fire, equipment wear, etc. All the collected data is transmitted to the ground control platform in real time through a wireless communication system for data analysis and processing. The AI system automatically analyzes the temperature change trend, equipment status, and potential faults, issues early warning information in a timely manner, and issues inspection instructions through the cluster management platform to control the drone to re-adjust its flight path and focus on checking areas with abnormal signs. In addition, all data and reports during the inspection process are summarized through the platform to generate a complete inspection report, providing a scientific basis for the maintenance and safety management of workshop equipment. Through this fully automated and unmanned inspection system, the coverage, efficiency, and accuracy of the inspection can be significantly improved, ensuring the production safety of the ore dressing workshop and the continuous and stable operation of equipment.

[0004] The existing technology has the following deficiencies:

[0005] In the ore dressing workshop, when the drone takes an inclined-angle shot of equipment near windows or reflective metal areas, affected by natural oblique light or the reflection of workshop lighting, local strong shadow areas that are extremely similar to the gray-scale characteristics of cracks will appear on the equipment surface; under the specific conditions of such oblique light irradiation, the shadow direction often tends to be consistent with the actual crack direction, and there is a certain degree of fuzzy diffusion at the shadow edge, thus interfering with the judgment of the edge structure in the image and causing visual confusion between the shadow and the crack in the image. Since such shadow areas usually have obvious gray-scale mutation characteristics in the image, the existing drone-based inspection technology for ore dressing workshops cannot dynamically adjust the judgment of the crack recognition path direction according to the degree of shadow gray-scale disturbance under oblique light irradiation, and still uses static image contour features for processing, easily misidentifying the shadow as the crack extension path, resulting in an offset in the crack recognition position or an error in the direction judgment. Furthermore, this will not only cause the crack path in the inspection result to deviate from the real position, misleading the equipment risk level assessment, but also affect the accuracy of subsequent modeling analysis of the crack development trend, inspection path planning, and maintenance scheduling, leading to problems such as false inspection, missed inspection, or repeated detection, reducing the overall accuracy and reliability of the inspection.

[0006] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] The object of the present invention is to provide a method for inspecting a beneficiation workshop based on an unmanned aerial vehicle to solve the problems in the above background art.

[0008] To achieve the above object, the present invention provides the following technical solution: A method for inspecting a beneficiation workshop based on an unmanned aerial vehicle, specifically including the following steps:

[0009] In a beneficiation workshop, when the unmanned aerial vehicle collects images of equipment at an inclined angle, according to the collected images and their corresponding attitude information, it is judged whether the images collected by the unmanned aerial vehicle will be affected by oblique light irradiation;

[0010] When the images collected by the unmanned aerial vehicle are affected by oblique light irradiation, identify the areas in the images with shadow interference characteristics and evenly divide them into several sub-areas;

[0011] Extract the image perturbation feature information of each sub-area from the images collected by the unmanned aerial vehicle and analyze it to determine the shadow gray-scale perturbation degree and perturbation type of each sub-area;

[0012] According to the perturbation types of each sub-area, respectively execute the judgment strategies for the walking directions of the corresponding crack recognition paths and perform dynamic regulation;

[0013] After completing the dynamic regulation of the crack recognition paths of each sub-area, perform fusion processing on the recognition results, output the complete crack path and its corresponding perturbation level label, and generate an inspection task review list according to the perturbation levels of each sub-area and write it into the scheduling list for subsequent reshooting and path optimization.

[0014] Preferably, when the images collected by the unmanned aerial vehicle are affected by oblique light irradiation, identifying the areas in the images with shadow interference characteristics specifically means: when the images collected by the unmanned aerial vehicle are affected by oblique light irradiation, by analyzing the gray-scale values of the collected images, detecting the brightness changes in all areas of the images, and combining the distribution of the reflected light sources in the images, identify the areas in the images with shadow interference characteristics, and the shadow interference characteristic areas include areas with blurred boundaries, brightness changes, and gray-scale change areas consistent with the light source direction;

[0015] And evenly divide it into several sub-areas, specifically: according to the spatial position and size of the shadow interference characteristic area in the image, use a grid algorithm to evenly divide the shadow interference area into multiple sub-areas, and the grid algorithm dynamically adjusts the division of the sub-areas by evenly distributing grid points in the shadow area according to the actual area and shape of the shadow area.

[0016] Preferably, image disturbance feature information of each sub-region is extracted from the images collected by the drone, and analyzed to determine the shadow gray-scale disturbance degree and disturbance type of each sub-region, specifically including the following steps:

[0017] Extract image disturbance feature information of each sub-region from the images collected by the drone, and perform preprocessing after extraction;

[0018] Extract gray-scale change feature information and edge blur feature information from the image disturbance feature information of each preprocessed sub-region, and analyze them after extraction to generate the shadow disturbance intensity coefficient and edge blur degree index of each sub-region respectively;

[0019] Based on the generated shadow disturbance intensity coefficients and edge blur degree indices of each sub-region, construct a shadow gray-scale disturbance evaluation model, and generate the shadow gray-scale disturbance coefficient of each sub-region through weighted summation;

[0020] Determine the preset shadow gray-scale disturbance coefficient threshold interval, and compare it with the generated shadow gray-scale disturbance coefficients of each sub-region after determination. Evaluate the shadow gray-scale disturbance degree of each sub-region according to the comparison result, and determine the disturbance type of each sub-region according to the evaluation result.

[0021] Preferably, the acquisition logic of the shadow disturbance intensity coefficient of each sub-region is as follows:

[0022] Extract gray-scale change feature information from the image disturbance feature information of each preprocessed sub-region, specifically including the gray-scale value of each pixel point in each sub-region of the image collected by the drone and the gray-scale mean value of all regions adjacent to each sub-region, and mark them respectively as and GN i , represents the gray-scale value of the m-th pixel point in the i-th sub-region of the image collected by the drone, and GN i represents the gray-scale mean value of all regions adjacent to the i-th sub-region in the image collected by the drone, i = 1, 2, 3,..., k, m = 1, 2, 3,..., j, and both k and j are positive integers;

[0023] Calculate the average value GPM of the gray-scale values of all pixel points in each sub-region of the image collected by the drone i , according to the formula:

[0024] Calculate the difference between the gray-scale value of each pixel point in each sub-region of the image collected by the drone and the average value of the gray-scale values of all pixel points in that sub-region According to the formula:

[0025] Determine the optimal values of the first adjustment factor γ and the second adjustment factor δ from the actual image data through a data fitting algorithm. The first adjustment factor γ is used to control the response sensitivity of the gray-scale fluctuation within the sub-region to the perturbation coefficient, and the second adjustment factor δ is used to adjust the response amplitude of the gray-scale difference between the sub-region and its neighborhood.

[0026] Calculate the shadow perturbation intensity coefficient of each sub-region. The specific calculation formula is as follows:

[0027]

[0028] In the formula, SDIC i is the shadow perturbation intensity coefficient of the i-th sub-region.

[0029] Preferably, the acquisition logic of the edge blur degree index of each sub-region is as follows:

[0030] Extract the edge blur feature information from the image perturbation feature information of each pre-processed sub-region, specifically including the edge gradient value and the gray-scale value of each edge pixel point within each sub-region in the image collected by the drone, and respectively calibrate them as and represents the edge gradient value of the n-th edge pixel point within the i-th sub-region in the image collected by the drone, represents the gray-scale value of the n-th edge pixel point within the i-th sub-region in the image collected by the drone, where i = 1, 2, 3,..., k, n = 1, 2, 3,..., h, and both k and h are positive integers;

[0031] Calculate the average gray-scale value BGP i of all edge pixel points within each sub-region in the image collected by the drone. According to the formula:

[0032] Calculate the difference between the gray-scale value of each edge pixel point within each sub-region in the image collected by the drone and the average gray-scale value of all edge pixel points within this sub-region According to the formula:

[0033] Determine the optimal value of the exponential adjustment factor α from the actual image data through a data fitting algorithm. The adjustment factor α is used to control the response enhancement degree of the edge gradient to the edge blur degree index;

[0034] Calculate the edge blur degree index of each sub-region. The specific calculation formula is as follows:

[0035]

[0036] wherein, EBI i is the edge blur index of the i-th sub-region.

[0037] Preferably, based on the shadow perturbation intensity coefficient SDIC i and the edge blur index EBI i of each generated sub-region, a shadow gray-scale perturbation evaluation model is constructed, and the shadow gray-scale perturbation coefficient of each sub-region is generated by weighted summation. The specific calculation formula is as follows:

[0038] SGDC i = ω1 * SDIC i + ω2 * EBI i

[0039] wherein, SGDC i is the shadow gray-scale perturbation coefficient of the i-th sub-region, ω1 and ω2 are non-zero weight coefficients of the shadow perturbation intensity coefficient SDIC i and the edge blur index EBI i of each sub-region, respectively, and ω1 + ω2 = 1.

[0040] Preferably, a preset shadow gray-scale perturbation coefficient threshold interval [SGDC min , SGDC max is determined, and after determination, it is compared with the shadow gray-scale perturbation coefficient SGDC i of each generated sub-region. According to the comparison result, the shadow gray-scale perturbation degree of each sub-region is evaluated, and according to the evaluation result, the perturbation type of each sub-region is determined. The specific comparison analysis is as follows:

[0041] If SGDC i < SGDC min , the shadow gray-scale perturbation degree of this sub-region is a low perturbation degree, and the perturbation type of this sub-region is a weak perturbation region;

[0042] If SGDC min ≤ SGDC i ≤ SGDC max , the shadow gray-scale perturbation degree of this sub-region is a medium perturbation degree, and the perturbation type of this sub-region is a medium perturbation region;

[0043] If SGDC i > SGDC max , the shadow gray-scale perturbation degree of this sub-region is a high perturbation degree, and the perturbation type of this sub-region is a strong perturbation region.

[0044] Preferably, according to the perturbation type of each sub-region, a corresponding crack recognition path trend judgment strategy is respectively executed, specifically including:

[0045] For the sub-region with weak disturbance type, the specific strategy for judging the trend of the crack recognition path is as follows: adopt the default benchmark path recognition algorithm, directly perform linear fitting based on the gray mutation edge in the image, extract the main crack direction, and extend the crack boundary according to the shortest path principle, without performing redundant filtering, angle correction, and auxiliary image fusion processing;

[0046] For the sub-region with medium disturbance type, the specific strategy for judging the trend of the crack recognition path is as follows: on the basis of performing the default path recognition algorithm, superimpose the edge fusion calculation based on multi-angle image acquisition, perform curvature compensation and node smoothing on the discontinuity breakpoints and direction mutations that appear in the path extraction, and adjust the path through the historical recognition trajectory;

[0047] For the sub-region with strong disturbance type, the specific strategy for judging the trend of the crack recognition path is as follows: preferentially call the anti-disturbance enhanced recognition mode, perform background enhancement, contrast adjustment, and artifact suppression processing on the image, and apply the high-order edge reconstruction algorithm based on model learning in the path judgment process, combined with the recognition results of adjacent non-disturbed regions, to perform fitting backtracking and abnormal elimination on the recognition path;

[0048] The specific method of dynamic regulation is as follows: continuously monitor the disturbance type of each sub-region during the generation of the recognition path, switch the corresponding path judgment strategy according to the disturbance type, and introduce a disturbance factor to participate in the path weight calculation during the path generation process to adjust the path trend priority in real time; if the recognition confidence of a certain sub-region is lower than the set threshold, mark this path segment as a "path to be rechecked" and synchronously write it into the reshooting schedule list.

[0049] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0050] 1. By constructing an image oblique light irradiation influence determination mechanism and a shadow interference feature region recognition process, the present invention can accurately identify local strong shadow regions caused by oblique shooting and natural or artificial oblique light sources, significantly enhancing the system's perception ability of image disturbance sources under complex lighting conditions. Further, by performing grid division on this shadow region, a finer-grained spatial modeling of image disturbance is realized, providing stable and hierarchical disturbance information support for the subsequent path recognition strategy, solving the problems of rough positioning of disturbance sources and single processing methods in traditional technologies, and improving the analysis basic quality of inspection data from the source.

[0051] 2. The present invention constructs a complete mathematical model for gray-scale perturbation analysis by defining two advanced feature parameters, namely, "shadow perturbation intensity coefficient" and "edge blur index", which can be calculated in real time based on image data. Through weighted summation, a unified "shadow gray-scale perturbation coefficient" is generated, and finally, the accurate classification of image perturbation levels is achieved. This solution not only integrates two key interference dimensions, namely local gray-scale fluctuations and edge sharpness, but also has an adjustable and optimizable model adaptability, supporting flexible adaptation to image characteristics under different devices, materials, and lighting conditions. By introducing a perturbation coefficient threshold interval for perturbation type classification, the system can intelligently perceive the risk level of the image recognition area, significantly improving the context adaptation ability of crack recognition and the accuracy of classification judgment.

[0052] 3. Based on the perturbation recognition results, the present invention introduces three different crack recognition path strategies, corresponding to weak, medium, and strong perturbation regions respectively. Combined with the dynamic strategy switching mechanism and the confidence monitoring feedback mechanism, it can adaptively adjust the algorithm process and judgment logic of the crack recognition path for different image perturbation environments. Especially in strong perturbation scenarios, by enhancing the recognition mode, edge reconstruction, and path fitting and complementing, etc., problems such as path drift, breakage, or deviation caused by shadow misjudgment are effectively overcome. Finally, through perturbation level annotation and path fusion, a complete and reliable crack path map is output. The system can also generate a recheck task list and write it into the scheduling list, forming a task closed-loop for automatic reshooting and subsequent optimization, thereby comprehensively improving the accuracy, robustness, and full-process closed-loop processing ability of inspection and recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0054] Figure 1 It is a schematic flow chart of a method for inspecting a beneficiation workshop based on an unmanned aerial vehicle according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that this disclosure will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0056] The present invention provides a method for inspecting a beneficiation workshop based on an unmanned aerial vehicle as shown in Figure 1 and specifically includes the following steps:

[0057] In the ore dressing workshop, when the UAV collects images of equipment at an inclined angle, it determines whether the images collected by the UAV are affected by oblique light illumination based on the collected images and their corresponding attitude information;

[0058] To determine whether the image is affected by oblique light illumination, it can be achieved by combining image analysis and attitude data. First, the software extracts the illumination information in the image captured by the UAV, that is, by analyzing the distribution of light sources in the image to detect whether there is strong light illumination or reflection. This can be achieved through the analysis of the brightness and contrast of the image. For example, by calculating the gray value distribution of the bright areas in the image to determine whether there are local strong light or reflection areas. Next, the software makes a further judgment based on the flight attitude information of the UAV (such as flight angle, shooting angle, and fuselage attitude), combined with the illumination direction. The sensor data of the UAV (such as gyroscope, accelerometer) can provide the angle information of the equipment relative to the ground, and then through a mathematical model, these information are combined with the brightness change of the image to infer the possibility of the image being affected by oblique light illumination. If the bright mutation area in the image matches the shooting angle of the UAV and the light source direction, it can be considered that the image is affected by oblique light illumination.

[0059] By combining the image data and the attitude information of the UAV to determine whether the image is affected by oblique light illumination, it can effectively improve the recognition accuracy of oblique light interference. The strong light or reflection areas in the image are often very similar to the crack features and are easily misidentified as cracks, resulting in deviation of the subsequent recognition path. If only relying on static image features or only using flight data for judgment, it is easy to miss the complex illumination changes in the image, resulting in misjudgment. By combining the two methods, it can more accurately determine which image areas are affected by oblique light interference, thereby providing a more reliable basis for subsequent image processing, avoiding unnecessary misidentifications, and improving the accuracy of crack recognition and the reliability of inspection results. This method can enhance the intelligence and adaptability of the inspection process and ensure the inspection effect of the UAV under complex illumination conditions.

[0060] When the images collected by the UAV are affected by oblique light illumination, identify the areas in the images with shadow interference features and evenly divide them into several sub-areas;

[0061] In this embodiment, when the images collected by the UAV are affected by oblique light illumination, the areas in the images with shadow interference features are identified specifically as follows: when the images collected by the UAV are affected by oblique light illumination, by analyzing the gray value of the collected images, detecting the brightness changes in all areas of the images, and combining the distribution of the reflected light sources in the images, the areas in the images with shadow interference features are identified. The shadow interference feature areas include areas with blurred boundaries, brightness changes, and gray value change areas consistent with the light source direction;

[0062] To detect the area with shadow interference features in an image, the brightness changes and the distribution of reflected light sources in the image can be analyzed through image processing algorithms combined with machine learning models. The specific implementation steps are as follows: First, the software can detect the brightness change areas in the image by analyzing the gray values of the image. By calculating the gray value of each pixel in the image and using image difference techniques (such as Sobel operator or Laplacian operator) to detect the sudden change of gray scale, the areas with obvious brightness changes in the image are highlighted. Then, the software analyzes the distribution of light sources in the image, combines the shooting angle of the drone and the light source position to identify the reflected light sources or local strong light areas in the image. This process can be achieved through the correlation analysis between the light source direction and the image brightness. The software uses the attitude information provided by the flight control system to infer the light source in the image, and then determines the brightness change areas consistent with the light source direction. For these areas, the software will further extract features, such as boundary blur degree, brightness change amplitude, and gray scale change consistent with the light source direction. These features are significant signs of shadow interference. The purpose of doing this is to accurately identify the shadow interference areas by combining gray scale changes, light source distribution, and flight angle. Traditional image processing methods often cannot distinguish the gray scale features of shadows and cracks, which easily leads to misidentification. By combining the flight attitude and the light direction, it can accurately judge which areas are interfered by oblique light, thus reducing the confusion between the shadow area and the crack path and improving the accuracy of image analysis. Finally, through the automated analysis of the software, more reliable data support can be provided for subsequent path recognition and dynamic regulation, ensuring the efficiency and accuracy of drone inspection.

[0063] And evenly divide it into several sub-regions. Specifically: According to the spatial position and size of the shadow interference feature area in the image, use the grid algorithm to evenly divide the shadow interference area into multiple sub-regions. The grid algorithm evenly distributes grid points within the shadow area and dynamically adjusts the division of sub-regions according to the actual area and shape of the shadow area to ensure that the size of each sub-region is suitable for subsequent analysis and processing.

[0064] In order to evenly divide the shadow interference area in the image into multiple sub-regions, the division strategy can be dynamically adjusted through a grid algorithm combined with image analysis and shape adaptability adjustment. First, the software detects the spatial position and size of the shadow interference area to obtain the size and shape characteristics of this area. Next, the grid division algorithm is adopted to evenly distribute grid points within the shadow area, and these grid points can be adjusted through an automatic adaptation method based on the area and shape of the region. Specifically, the size and shape of the grid will be adaptively adjusted according to the actual size and shape of the shadow area. For a larger shadow area, the size of the grid will become larger; while for a smaller or irregularly shaped area, the grid will be refined to ensure that the area of each sub-region is suitable for subsequent analysis and processing. To ensure the accuracy of grid division, the software can adopt an adaptive algorithm based on the region boundary. By analyzing the boundary shape of the shadow area, the shape and distribution density of the grid are dynamically adjusted. For a shadow area with an irregular boundary, the software will automatically generate a grid of irregularly shaped sub-regions to adapt to different morphological shadow interference areas. The division of each sub-region is finely adjusted according to the actual position and size of the shadow area to ensure that the size of each sub-region is suitable for subsequent image processing tasks such as feature extraction and interference evaluation.

[0065] The purpose of this is to accurately adapt to shadow areas of different sizes and shapes by dynamically adjusting the size and distribution of the grid, thereby ensuring the accuracy of subsequent analyses (such as shadow interference analysis and crack path recognition). A unified grid division strategy can ensure that each sub-region has sufficient analysis data, avoid omission or uneven processing, improve the meticulousness and accuracy of image analysis, and ultimately enhance the overall effectiveness and reliability of UAV inspection.

[0066] Extract the image perturbation feature information of each sub-region from the images collected by the UAV, and analyze it to determine the shadow gray-scale perturbation degree and perturbation type of each sub-region;

[0067] In this embodiment, extracting the image perturbation feature information of each sub-region from the images collected by the UAV and analyzing it to determine the shadow gray-scale perturbation degree and perturbation type of each sub-region specifically includes the following steps:

[0068] Extract the image perturbation feature information of each sub-region from the images collected by the UAV and perform preprocessing after extraction;

[0069] Extracting image perturbation feature information can be achieved through image processing techniques and feature extraction algorithms. First, the software can divide the image into multiple sub-regions through image segmentation technology, and each sub-region represents a possible shadow interference area. Then, gray value analysis is adopted, such as calculating the mean value, standard deviation, and gray change amplitude of pixel gray values within each sub-region. These statistics can effectively reflect the illumination differences and gray changes in the sub-regions, thereby helping to identify the existence of shadow areas. In addition to gray analysis, edge detection algorithms (such as Sobel operator, Canny edge detection) can extract the edge information of the sub-regions to help identify the boundaries between shadow areas and the background in the image. Finally, by combining local brightness changes (i.e., the difference between each pixel and the regional mean value), the features of shadows and illumination changes in the image can be further enhanced, thus providing more accurate perturbation feature information.

[0070] Image preprocessing is to improve the accuracy of subsequent feature extraction and analysis, remove noise, and improve image quality. Common preprocessing steps include noise removal and image enhancement. During the image acquisition process, noise, blur, and uneven illumination may affect the extraction of perturbation features. Therefore, Gaussian filtering or median filtering is first used to remove random noise in the image, which helps to smooth the image and retain important edge information. Next, histogram equalization or contrast enhancement techniques are used to improve the contrast of the image, making the gray changes in the shadow areas more obvious and enhancing the detail expression of the image. For shadow interference in complex backgrounds, image normalization processing can reduce the differences brought by different illumination conditions, making the perturbation features in the shadow areas more prominent and facilitating the subsequent calculation of the perturbation intensity and blur degree index values. The preprocessed image will appear clearer and more consistent, providing a more reliable data basis for subsequent feature extraction, analysis, and classification.

[0071] Extract the gray change feature information and edge blur feature information from the image perturbation feature information of each preprocessed sub-region, and perform analysis after extraction to generate the shadow perturbation intensity coefficient and edge blur degree index for each sub-region respectively;

[0072] To extract the grayscale change feature information and edge blur feature information from the image perturbation feature information of each preprocessed sub-region, the following methods can be used: First, the grayscale change feature information can be achieved through local grayscale difference analysis. Specifically, the software can calculate the difference between each pixel point in each sub-region and the grayscale mean value of the sub-region to obtain the grayscale change amount, which reflects the grayscale deviation degree of each pixel relative to the mean value within the sub-region. By calculating the difference metrics (such as standard deviation, mean difference) of each pixel to quantify the local grayscale change, the grayscale mutation regions caused by shadows or reflections in the image can be effectively captured. In addition, the edge blur feature information can be extracted through edge detection algorithms (such as Sobel operator, Canny operator, etc.). The edge detection algorithm can calculate the grayscale gradient of each sub-region in the image and identify the edge contours of the shadow region. Subsequently, through blur measurement, such as calculating the grayscale change rate and blur degree of the edge, the software can quantify the blur degree of the shadow region and reflect the edge sharpness of the shadow region. In this way, the quantitative feature information of grayscale change and edge blur in the image can be extracted simultaneously, providing key data support for subsequent perturbation analysis and recognition.

[0073] Based on the generated shadow perturbation intensity coefficients and edge blur degree indices of each sub-region, construct a shadow grayscale perturbation evaluation model, and generate the shadow grayscale perturbation coefficients of each sub-region through weighted summation;

[0074] Determine the pre-set threshold interval of the shadow grayscale perturbation coefficient, and compare it with the generated shadow grayscale perturbation coefficients of each sub-region after determination. Evaluate the shadow grayscale perturbation degree of each sub-region according to the comparison result, and determine the perturbation type of each sub-region according to the evaluation result.

[0075] To determine the pre-set threshold interval of the shadow grayscale perturbation coefficient, data analysis and statistical modeling can be used. First, the software can perform data clustering analysis based on a large amount of actually collected image data. For example, the K-means clustering algorithm or Gaussian Mixture Model (GMM) can be used to analyze different types of perturbation coefficient data. These algorithms can classify the data into different categories according to different degrees of perturbation features in the image (such as shadow intensity, edge blur, etc.) and calculate the perturbation coefficient distribution of each category. In this way, the software can identify the typical values of different perturbation types and calculate the corresponding threshold interval based on these values. In addition, statistical analysis methods such as mean and standard deviation analysis can be used to determine the normal range of the perturbation coefficient. For example, the software can calculate the mean and standard deviation of the shadow grayscale perturbation coefficient in the training dataset, and then set the threshold interval as the mean plus or minus several standard deviations, which can cover the perturbation coefficient values in most normal situations. For images with high-intensity or abnormal perturbations, the software can set specific threshold intervals to handle abnormal situations by analyzing these data separately. Through this method, the software can dynamically adjust the threshold interval to adapt to different lighting conditions and environmental changes, ensuring accurate evaluation in complex environments.

[0076] In this embodiment, the acquisition logic of the shadow perturbation intensity coefficient of each sub-region is as follows:

[0077] Extract the grayscale change feature information from the image perturbation feature information of each pre-processed sub-region, specifically including the grayscale value of each pixel point in each sub-region of the image collected by the drone and the grayscale mean value of all regions adjacent to each sub-region, and mark them respectively as and GN i , represents the grayscale value of the m-th pixel point in the i-th sub-region of the image collected by the drone, and GN i represents the grayscale mean value of all regions adjacent to the i-th sub-region in the image collected by the drone, where i = 1, 2, 3,..., k, m = 1, 2, 3,..., j, and both k and j are positive integers;

[0078] To achieve real-time acquisition of two types of data, namely, "the gray values of each pixel point in each sub-region of the images collected by the drone" and "the average gray value of all regions adjacent to each sub-region", it can be accomplished through software image processing methods such as image tiling, sliding window, and matrix operations. First, after the software receives the images taken by the drone, it divides the images into several regular or adaptive sub-regions through a preset grid strategy or dynamic region segmentation algorithm, and maps each sub-region into a two-dimensional gray matrix. Subsequently, the system uses the image gray channel extraction technology to traverse the gray matrix of each sub-region and extract the gray value of each pixel one by one, thereby forming a complete data set of "the gray values of each pixel point in the sub-region", which can be used to reflect the internal brightness distribution and gray fluctuations of the sub-region. Then, after the software extracts the coordinate information of each sub-region, it expands a certain pixel range outward based on the boundary position of the region, uses the sliding window algorithm to obtain all adjacent sub-regions or adjacent pixel sets, and calculates the average of all their gray values, thereby obtaining the corresponding "average gray value of the adjacent region". This average value can reflect the comparison benchmark of the sub-region under the image background or surrounding lighting conditions, which helps to determine whether the region is affected by shadow interference due to sudden gray changes in the subsequent process. This process can be automatically executed by the image processing engine in units of frames after the image is loaded, so as to achieve real-time calculation and dynamic update during the drone flight, not only with fast response, but also able to adapt to the inspection requirements under continuous image data.

[0079] Calculate the average gray value GPM of all pixel points in each sub-region of the images collected by the drone i , according to the formula:

[0080] Calculate the difference between the gray value of each pixel point in each sub-region of the images collected by the drone and the average gray value of all pixel points in the sub-region According to the formula:

[0081] Determine the optimal values of the first adjustment factor γ and the second adjustment factor δ from the actual image data through a data fitting algorithm. The first adjustment factor γ is used to control the response sensitivity of the internal gray fluctuations of the sub-region to the perturbation coefficient, and the second adjustment factor δ is used to adjust the response amplitude of the gray difference between the sub-region and its neighborhood;

[0082] In order to determine the optimal values of the first adjustment factor and the second adjustment factor, the data fitting algorithm can be executed by the software to perform parameter training and optimization based on a large amount of actual inspection image data. The specific implementation method is as follows: First, a training image set containing a variety of typical shadow disturbance situations is constructed, and the sub-regions marked in each image are manually or semi-automatically assigned with real disturbance level labels (such as weak disturbance, medium disturbance, strong disturbance) as the "reference standard" for fitting. Subsequently, the software calculates its shadow disturbance intensity coefficient (SDIC) according to the current formula on each sub-region, and continuously adjusts the parameter combination by adjusting the first adjustment factor (controlling the response amplification degree of the grayscale fluctuation inside the pixel) and the second adjustment factor (controlling the overall grayscale difference between the region and the neighborhood in the disturbance assessment. The weight of the influence is minimized. This process can be completed by using the minimum mean square error (MSE) regression or gradient descent optimization algorithm to complete the fitting, so that the final selected factor combination has the minimum prediction error under the full sample. The reason for introducing these two adjustable factors is that there is a nonlinear relationship between the grayscale change characteristics of different areas in the image and the degree of interference from their real shadows. Fixed calculation formulas are often unable to adapt to complex and diverse lighting conditions and material reflection characteristics. The personalized factor combination trained by fitting can enhance the adaptability and discrimination ability of the model, making it more accurate and stable in identifying weak and strong disturbances in practical applications, thereby improving the versatility and reliability of the entire inspection system in multiple scenarios. The entire process can be automatically executed in the image analysis platform without manual intervention, thus meeting the needs of real-time processing and batch modeling.

[0083] Calculate the shadow disturbance intensity coefficient of each sub-area. The specific calculation formula is as follows:

[0084]

[0085] Where, SDIC i is the shadow disturbance intensity coefficient of the i-th sub-region.

[0086] The reason why this formula is used to calculate the shadow disturbance intensity coefficient of each sub-region is to comprehensively quantify the dual effects of the grayscale fluctuation inside the sub-region and the illumination contrast of the surrounding area on the shadow disturbance intensity, and to enhance the distinguishability of the disturbance characteristics through exponential and logarithmic operations. Specifically, the numerator in the formula is calculated by calculating the difference between the grayscale value of each pixel and the grayscale mean of its sub-region. The power operation is performed to control the weight amplification relationship. The first adjustment factor γ is used to control the response sensitivity of the grayscale fluctuation inside the sub-region to the disturbance coefficient, so as to achieve sensitive capture of local subtle shadow disturbances. The denominator extracts the absolute value of the difference between the grayscale mean of the sub-region and the grayscale mean of its adjacent region, performs an exponential operation, and then compresses and adjusts it through a logarithmic function. Its function is to enhance the discrimination of the brightness difference between the sub-region and the surrounding region on the disturbance intensity, while avoiding the numerical instability caused by extreme values. The second adjustment factor δ controls the nonlinear enhancement degree of the illumination difference on the overall disturbance evaluation. The entire calculation logic ensures that the evaluation has normalized characteristics by averaging the disturbance value of each pixel (divided by the number of pixels j), so that the disturbance coefficients of regions of different sizes are comparable, thereby providing a reliable, stable and adjustable evaluation basis for the subsequent quantitative division and dynamic regulation of the degree of shadow disturbance.

[0087] The shadow disturbance intensity coefficient SDIC of the i-th sub-area i The value of directly reflects the degree to which the sub-region is disturbed by the shadow in the image, and is the core quantitative indicator for evaluating the degree of shadow grayscale disturbance. Specifically, if SDIC i If the value of is large, it means that the grayscale fluctuation of the pixels in the sub-region is more intense, and there is a significant brightness difference with the adjacent regions, indicating that the region is significantly disturbed by shadows and belongs to a strong disturbance region. On the contrary, if SDIC i The smaller the value, the more uniform the grayscale distribution inside the sub-region is, and the brightness contrast between the sub-region and the neighboring region is not strong, indicating that the shadow effect is weak and it belongs to the low disturbance area. Therefore, by setting a set of perturbation coefficient threshold intervals with discrimination, the SDIC i The value range of is used to accurately evaluate the degree of shadow grayscale disturbance of the ith sub-region, and classify it as a weak disturbance, medium disturbance or strong disturbance area, which serves as the judgment basis for the subsequent crack identification path regulation.

[0088] In this embodiment, the logic for obtaining the edge blur index of each sub-region is as follows:

[0089] The edge blur feature information is extracted from the preprocessed image disturbance feature information of each sub-region, specifically including the edge gradient value of each edge pixel point in each sub-region of the image collected by the drone and the grayscale value of each edge pixel point, and calibrated as and It represents the edge gradient value of the nth edge pixel in the ith sub-region of the image collected by the drone. represents the gray value of the nth edge pixel in the ith sub-region of the image collected by the drone, i = 1, 2, 3, ..., k, n = 1, 2, 3, ..., h, k and h are both positive integers;

[0090] To achieve real-time acquisition of two types of data, namely, "edge gradient values of each edge pixel point in each sub-region" and "gray values of each edge pixel point", automatic extraction and analysis can be performed on the image data stream after the UAV image acquisition through the image processing algorithms integrated in the software. Specifically, first, the software divides the image collected by the UAV into sub-regions (such as based on grids, sliding windows, or dynamic region segmentation algorithms), and uses edge detection operators (such as Sobel, Prewitt, or Canny) to perform edge extraction processing on the image data in each sub-region. This process calculates the gradient value of each edge pixel point, that is, the edge gradient value This value reflects the change rate of the pixel gray value in the horizontal and vertical directions. The larger the gradient, the more likely it is that the point is at the structural edge or the brightness mutation, and it has a strong guiding effect on the crack recognition path. At the same time, the software can directly read the original pixel values of these edge pixel points in the gray channel, that is, the gray values It represents the numerical size of the point in the image brightness space and is the basic parameter for determining the edge light and dark contrast and brightness continuity. To adapt to the changes in the images during the continuous flight of the UAV, the system will input the image frames into the processing engine frame by frame, and extract these pixel-level feature information in real time and cache it in each frame of the image, so that the edge gradient value and the gray value can dynamically reflect the local illumination and structural changes in the current inspection scene, and then support the subsequent calculation of the edge blur index and the recognition of the shadow perturbation level. The whole process is highly automated at the software level without manual intervention, and can run stably during the flight to ensure the continuity and real-time of information extraction.

[0091] Calculate the mean gray value BGP of all edge pixel points in each sub-region of the image collected by the UAV i , according to the formula:

[0092] Calculate the difference between the gray value of each edge pixel point in each sub-region of the image collected by the UAV and the mean gray value of all edge pixel points in that sub-region According to the formula:

[0093] Determine the optimal value of the exponential adjustment factor α from the actual image data through a data fitting algorithm. The adjustment factor α is used to control the response enhancement degree of the edge gradient to the edge blur index;

[0094] To achieve the optimal value of the exponential adjustment factor, the software can automatically learn and optimize the value of this factor from a large amount of actual image data through a data fitting algorithm. Specifically, the software will first extract the edge gradient values and brightness change amounts of each sub-region from the images collected by the drone, and calculate the preliminary edge blur index. Then, by comparing the actual edge blur of these images with the "actual disturbance degree" calibrated manually, the software can use data fitting algorithms such as the least mean square error (MSE) or gradient descent method to gradually adjust the value of the adjustment factor α to minimize the error between the predicted edge blur and the actual calibrated value. This process is achieved through continuous iteration. Each time the factor α is adjusted, the software will automatically adjust the parameters according to the current fitting error to find an optimal factor value that can best match the actual image data.

[0095] The reason for determining the optimal value of the exponential adjustment factor through this method is that the edge blurriness in the image is a highly dynamic and non-linear feature, which is not only affected by the edge gradient, but also closely related to factors such as lighting conditions, object surface materials, and shooting angles. A fixed adjustment factor may not be able to adapt to the changes in different image scenarios. Therefore, by optimizing this factor through data fitting, it can have strong adaptability in various scenarios, be able to more precisely control the influence of the edge gradient on the edge blur index, and thus improve the accuracy and stability of the edge blur index calculation. The entire process is automatically carried out in the image analysis software, which can achieve real-time learning and dynamic optimization, ensuring that in a changing environment, the algorithm can always execute tasks with the best parameters.

[0096] Calculate the edge blur index of each sub-region. The specific calculation formula is as follows:

[0097]

[0098] In the formula, EBI i is the edge blur index of the i-th sub-region.

[0099] The reason for using this calculation formula to obtain the edge blur index EBI of each sub-region i is to comprehensively quantify the coupling relationship between the edge sharpness and brightness change characteristics in the shadow region of the image, so as to accurately evaluate its interference degree on the image crack recognition path. Specifically, in the numerator of the formula, is used to extract the edge gradient value of each edge pixel point, reflecting the change intensity of the edge at this point. The exponential adjustment factor α further amplifies the sensitivity to high-gradient edges to highlight the dominant role of the edge clear region in the calculation; this gradient value is multiplied by , Indicates the difference between the grayscale value of this pixel point and the average grayscale value of the sub-region edge. After taking the logarithm, it can compress the response range of excessive grayscale changes while maintaining the influence of small changes, so that this item comprehensively reflects the fuzziness of the edge point; the denominator part constitutes an inverse proportional adjustment factor. When the brightness change is small (i.e., fuzzy), the value of this item is large, which plays a role in compressing the numerator, thereby enhancing the recognition ability of the fuzzy area; when the brightness change is large (i.e., the edge is clear), this item approaches 1, retaining the complete contribution of the numerator. Finally, the average is calculated in units of h to ensure the comparability of the calculation results for regions with different numbers of pixels. Overall, this formula realizes the quantitative and accurate recognition of the shadow fuzzy area through the non-linear combination of grayscale difference and edge strength, which is a key link in realizing stable crack detection under complex light interference.

[0100] The edge blurriness index EBI of the i-th sub-region i The numerical size of is closely related to the evaluation of the shadow grayscale perturbation degree of this region, specifically reflected as a quantitative reflection of the edge fuzziness of this region, thereby assisting in judging the strength of the shadow interference it receives. EBI i The larger the value, it indicates that the edge pixels in this sub-region generally have a relatively high edge gradient (i.e., the boundary structure is clear) and at the same time are accompanied by large fluctuations in brightness change, indicating that the edge characteristics of this region are unstable and are easily affected by oblique light shadow interference, so it is judged as a strongly perturbed region; on the contrary, when EBI i The value is small, indicating that the edge transition is gentle, fuzzy, and the grayscale distribution is uniform, belonging to a region with weak texture or continuous brightness, indicating that this sub-region is less affected by shadow interference and can be classified as a weakly perturbed region. Therefore, the edge blurriness index can not only reflect the edge manifestation form of grayscale perturbation, but also play a key role in assisting in identifying abnormal regions in a complex shadow environment, and is one of the core judgment bases for realizing highly reliable inspection image interpretation.

[0101] In this embodiment, based on the shadow perturbation intensity coefficient SDIC i and the edge blurriness index EBI i of each generated sub-region, a shadow grayscale perturbation evaluation model is constructed, and the shadow grayscale perturbation coefficient of each sub-region is generated by weighted summation. The specific calculation formula is as follows:

[0102] SGDC i =ω1*SDIC i +ω2*EBI i

[0103] In the formula, SGDC i is the shadow grayscale perturbation coefficient of the i-th sub-region, and ω1 and ω2 are the shadow perturbation intensity coefficient SDIC i and the edge blurriness index EBI of each sub-regioni with non - zero weight coefficients, and ω1 + ω2 = 1.

[0104] To calculate the Shadow Gray - scale Disturbance Coefficient (SGDC) of each sub - region i The system will, based on two acquired core parameters - the Shadow Disturbance Intensity Coefficient (SDIC) i and the Edge Blur Index (EBI) i , fuse these two indicators into a unified disturbance evaluation value by constructing a weighted evaluation model. In the specific implementation process, the software will simultaneously call the corresponding SDIC i and EBI i values for each sub - region, and introduce two non - zero weight coefficients ω1 and ω2, where ω1 represents the relative importance of SDIC i in the comprehensive evaluation, and ω2 represents the influence weight of EBI i . The sum of the two is always 1 to ensure the normalization and comparability of the results. These two weights can be optimized and set through a training set according to the actual application scenario, or can be preset as empirical values. For example, in a scenario with drastic lighting changes, increase the proportion of ω1 to enhance the response to gray - scale disturbance, and in a complex background with significant edge interference, increase the ω2 weight to pay more attention to the impact of blurred edges. Finally, the system completes the fusion calculation according to the formula SGDC i = ω1 * SDIC i + ω2 * EBI i , making the SGDC i of each sub - region a key quantitative indicator comprehensively reflecting the degree of its joint influence by shadow gray - scale and edge disturbance, providing a unified judgment basis for subsequent disturbance grading and path optimization. The whole process is completed in real - time through image analysis software to ensure the continuous output of highly reliable image interference evaluation results during the inspection process.

[0105] In this embodiment, a pre - set threshold interval of the shadow gray - scale disturbance coefficient [SGDC min , SGDC max is determined, and after determination, it is compared with the generated shadow gray - scale disturbance coefficient SGDC i of each sub - region. According to the comparison result, the shadow gray - scale disturbance degree of each sub - region is evaluated, and the disturbance type of each sub - region is determined according to the evaluation result. The specific comparison and analysis are as follows:

[0106] If SGDC i < SGDC min , the shadow gray - scale disturbance degree of this sub - region is a low - disturbance degree, and the disturbance type of this sub - region is a weak - disturbance region;

[0107] This situation means that this area shows low gray - level volatility and edge blurriness in the image. From the perspective of image features, the brightness distribution of such areas is relatively stable, the difference in internal gray - level values is small, and the edge structure is clear, without being significantly interfered by oblique light shadows, reflections, or high - contrast boundaries. Therefore, in the crack recognition task, this area has high image stability and interpretability, is suitable for directly implementing standard image recognition algorithms, has a high recognition result accuracy, and does not require additional intervention or post - processing, which can significantly reduce the misrecognition rate and subsequent processing costs.

[0108] If SGDC min ≤SGDC i ≤SGDC max , the degree of shadow gray - level perturbation of this sub - area is medium perturbation degree, and the perturbation type of this sub - area is medium - perturbation area;

[0109] This situation indicates that there is medium - degree perturbation interference in the image features of this sub - area. Usually, it is manifested as the gray - level change amplitude of this area is not smooth enough, the bright - dark transition is not natural enough, and there may be a certain degree of edge blurring or weak oblique light projection, but it has not reached the level of seriously affecting the recognition judgment. In actual inspection tasks, such areas need to introduce enhanced image recognition strategies or perform medium - degree fault - tolerance processing, such as fusing multi - angle image results, dynamic threshold adjustment, or soft - edge compensation methods, to improve recognition stability and accuracy, and prevent slight interference from causing path deviation or crack breakpoint judgment errors.

[0110] If SGDC i >SGDC max , the degree of shadow gray - level perturbation of this sub - area is high perturbation degree, and the perturbation type of this sub - area is strong - perturbation area.

[0111] This situation indicates that this area has been significantly interfered by image perturbations. Usually, it is manifested as drastic gray - level mutations, chaotic edge structures, severe shadow overlaps, or the existence of high - reflection areas, making it difficult to match the structural textures in the image with the actual surface features of the device. In this case, traditional crack recognition methods are prone to misjudgment, fracture, drift, or even missed detection, seriously affecting the accuracy of the inspection results. Therefore, such areas should be marked as high - risk image areas, included in the inspection re - check list, and priority should be given to reshooting, light reconstruction, path avoidance, or algorithm upgrade processing in subsequent scheduling to ensure the stability and intelligence of the overall inspection process.

[0112] According to the perturbation types of each sub - area, respectively execute the corresponding crack recognition path orientation judgment strategies and perform dynamic regulation;

[0113] In this embodiment, according to the perturbation types of each sub - area, respectively execute the corresponding crack recognition path orientation judgment strategies, specifically including:

[0114] For the sub-area where the disturbance type is the weak disturbance area, the crack identification path direction judgment strategy is as follows: the default reference path recognition algorithm is used to perform linear fitting directly based on the grayscale mutation edge in the image, the main direction of the crack is extracted, and the crack boundary is extended according to the shortest path principle, without the need to perform redundant filtering, angle correction and auxiliary image fusion processing;

[0115] In the sub-area where the disturbance type is a weak disturbance area, since the image grayscale distribution is uniform, the edge features are clear and the influence of light is small, the default reference path recognition algorithm can be directly applied by the software to efficiently extract the crack direction without image enhancement or multi-source information compensation. The specific implementation method is: first, the high grayscale change boundary in the image is extracted using an edge detection algorithm (such as Canny or Sobel), and the system performs connectivity analysis on these edge pixels to screen out linear crack candidate areas with coherence and ductility. Then, the edge set is linearly fitted by the least squares method or Hough transform to extract the dominant direction line segment of the crack; then, combined with the path extension logic, according to the directional change of the grayscale gradient in the image, the path is automatically extended along the main direction until the grayscale gradient of the crack edge tends to disappear or the structure is broken. In the whole process, there is no need to perform additional filtering and denoising, angle correction or cross-frame image fusion operations, which can greatly improve the computational efficiency. The reason for adopting this simplified process is that the weakly disturbed area itself has good image quality and structural clarity, and the crack information is highly recognizable in the original image. Executing the basic path extraction strategy can ensure the accuracy and stability of the recognition results, while saving computing resources and improving the real-time response capability of the system. It is particularly suitable for rapid judgment and on-the-fly judgment mission scenarios during UAV inspections.

[0116] For the sub-area with medium disturbance type, the crack identification path direction judgment strategy is as follows: on the basis of executing the default path recognition algorithm, the edge fusion calculation based on multi-angle image acquisition is superimposed, the discontinuity breakpoints and direction mutations appearing in the path extraction are subjected to curvature compensation and node smoothing, and the path is adjusted through the historical recognition trajectory to enhance the continuity and robustness of the path direction judgment;

[0117] In the sub-region with medium disturbance type, due to a certain degree of uneven transition or edge blur in the image grayscale, the crack path may be broken, offset or directionally jumped locally. Therefore, on the basis of the default path recognition algorithm, a multi-angle edge fusion and path repair mechanism needs to be superimposed to improve the continuity and robustness of path extraction. The specific implementation method is as follows: First, the system uses images of the same region taken by the drone from different angles during image acquisition, and adopts image registration and geometric correction methods to align the images from multiple perspectives; Subsequently, stable overlapping edge features from multiple angles are extracted through edge fusion algorithms (such as pixel consistency weighting or edge confidence map superposition), thereby enhancing the continuity of broken or blurred boundaries in the original image. For the remaining path breakpoints, the software further applies a path fitting algorithm based on the principle of minimum curvature to connect the broken segments and smooth the direction; At the same time, the recognized path in the historical inspection images is called, and through path matching and trajectory deviation analysis, the current path is corrected and supplemented to ensure the coherence of the trend judgment. The reason for adopting this strategy is that the image features in the medium disturbance region are between clear and severely disturbed, and although cracks can be detected but are unstable. If path deviations are not processed, it is easy to cause structural misjudgment or distortion of the extension trend. Therefore, through multi-angle information fusion and path callback mechanism, the adaptability of the recognition process can be enhanced while ensuring accuracy, so that the system can still maintain a reliable path judgment effect under complex disturbances.

[0118] For the sub-region with strong disturbance type, the specific strategy for judging the trend of the crack recognition path is as follows: First, call the anti-disturbance enhanced recognition mode, perform background enhancement, contrast adjustment and artifact suppression on the image, and apply a high-order edge reconstruction algorithm based on model learning during the path judgment process. Combining the recognition results of adjacent non-disturbed regions, the recognized path is fitted and backtracked and abnormal points are removed to ensure the recognition accuracy of the crack path in a strong disturbance environment;

[0119] In the sub - regions where the disturbance type is strong disturbance, images are often severely affected by shadow interference, light patches, metal reflections, or blurred superposition, etc., resulting in serious distortion of the crack edge information in the original image or fusion with the background. It is difficult for traditional edge detection or gray - scale analysis methods to stably identify the crack direction. Therefore, the software needs to switch to the disturbance - avoidance enhanced recognition mode first, and achieve reliable recognition of the crack direction through multi - step image enhancement and intelligent path back - propagation strategies. The specific implementation methods include: First, perform background enhancement and local contrast stretching on the image to make the original gray - scale distribution more balanced and form a separable gray - scale gradient between the crack region and the background region; at the same time, introduce an artifact suppression algorithm (such as morphological processing and high - frequency component suppression) to reduce the interference of reflections, highlights, or texture noise on the edge structure. Subsequently, the system calls a high - order edge reconstruction model trained based on deep learning or graph neural network to perform pixel - level edge prediction on the distorted region and reconstruct the potential crack edge direction. When generating the recognition path, the software will also reference the path results already identified in adjacent weak - disturbance or medium - disturbance regions, and perform path fitting and back - propagation completion through methods of direction extrapolation and boundary mapping. At the same time, identify abnormal nodes (such as sharp turning angles, pseudo - edge intersections) in the path and perform elimination processing. The reason for adopting such a highly intelligent processing strategy is that the strong - disturbance region essentially exceeds the fault - tolerance range of traditional image recognition algorithms. The extraction of crack information needs to be completed through the collaboration of context reasoning, global modeling, and interference repair means. Only by enhancing the image quality, fusing model predictions with adjacent region information can we ensure that there is still an available crack path judgment result under extreme interference conditions, and ensure the practicality and fault - tolerance of the entire inspection system in complex industrial scenarios.

[0120] The specific way of dynamic regulation is as follows: Continuously monitor the disturbance type of each sub - region during the process of generating the recognition path, switch the corresponding path judgment strategy according to the disturbance type, and introduce a disturbance factor to participate in the path weight calculation during the path generation process to adjust the path direction priority in real - time; if the recognition confidence of a certain sub - region is lower than the set threshold, mark this path segment as a "path to be re - verified", and synchronously write it into the reshooting scheduling list to ensure the whole - process dynamic optimization and stable output of the image recognition path.

[0121] To achieve the full-process dynamic regulation of the crack recognition path, the system can construct a disturbance perception mechanism and a strategy switching engine during the path generation process, monitor the disturbance types of sub-regions in real time at each stage of image processing, and flexibly adjust the path generation method and priority according to the preset strategy. The specific implementation method is as follows: After the system completes the calibration of the disturbance types of each sub-region (i.e., weak disturbance, medium disturbance, strong disturbance), it inputs them as dynamic parameters into the recognition process control module. This module makes a status judgment on each processing region in real time during the path recognition execution process and automatically calls the corresponding crack recognition path judgment strategy (such as benchmark recognition, fusion correction, or enhanced reconstruction). At the same time, the system sets a "disturbance factor" weight parameter for each sub-region, which is calculated based on its SGDC value and EBI value and is used to participate in path priority scheduling and path continuity optimization. For example, when splicing global paths, the path segments with smaller disturbance factors are preferentially connected, or in the path conflict judgment, the path direction of the region with a higher disturbance factor is suppressed. After the path is generated, the system also calculates the recognition confidence value for each path segment in combination with the pixel continuity, direction consistency, and edge feature consistency of the crack path. If this value is lower than the preset credible threshold, the corresponding path segment is automatically marked as a "path to be rechecked", and information such as its image number, coordinate position, and the disturbance level of the corresponding region is synchronously recorded and written into the reshooting schedule list as a reference for subsequent flight path optimization and secondary acquisition scheduling. The reason for adopting this dynamic regulation mechanism is that the disturbance degrees of different regions in the inspection images vary significantly, and a fixed path recognition process cannot be compatible with all scenarios. However, the dynamic strategy switching based on disturbance perception and the real-time adjustment of path weights can improve the overall recognition accuracy and the stability of path output. At the same time, through the "low-confidence reshooting" mechanism, an information closed-loop is realized to ensure the continuous operation reliability and high fault tolerance of the system under long-term inspection and complex working conditions.

[0122] After completing the dynamic regulation of the crack recognition paths of each sub-region, the recognition results are fused, and the complete crack path and its corresponding disturbance level label are output. Then, according to the disturbance levels of each sub-region, an inspection task review list is generated and written into the schedule list for subsequent reshooting and path optimization.

[0123] After completing the dynamic regulation of the crack recognition paths in each sub-region, to ensure the integrity of the inspection results and the intelligent scheduling of subsequent tasks, the system needs to execute a crack path fusion process and a disturbance level marking mechanism through software, and then generate a re-inspection list and a scheduling list to achieve the traceability of the recognition results and the closed-loop management of task execution. The specific implementation method is as follows: First, the system splices and fuses the sub-region recognition paths processed by their respective strategies in the order of spatial positions. Through algorithms such as edge connection matching, path direction consistency judgment, and node continuity scoring, it automatically repairs the small offsets or splicing breaks caused by strategy switching or interference boundaries between crack path segments, and outputs a complete and structurally coherent crack path. Subsequently, the system marks the corresponding disturbance level labels (such as weak disturbance, medium disturbance, strong disturbance) for each sub-region on this path, and establishes the mapping relationship between the path segments and the disturbance levels for subsequent analysis. Based on the above label data, the software automatically identifies all path segments classified as strong disturbance regions or path segments with recognition confidence lower than the threshold on this path, and generates a re-inspection list for the inspection according to metadata such as timestamp, image number, and geographical location information. The system records all items in this list into the task queue of the scheduling control module to form a structured "reshooting task list" and "path optimization suggestions". The reason for such a design is that during the actual inspection of the UAV, affected by factors such as environmental lighting, equipment material, and flight attitude, there may still be uncertainties in recognition in some regions even after dynamic strategy processing. Through the fusion of recognition results and the backtracking of disturbance levels, not only can a complete inspection path map be formed, but also it can ensure that potential risk regions are systematically re-inspected, ultimately improving the closed-loop efficiency, task accuracy, and system stability of the entire inspection task.

[0124] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by technicians in this field according to the actual situation.

[0125] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0126] It should be understood that in various embodiments of the present application, the sequence numbers of the above processes do not mean the order of execution, and the order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0127] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0128] In several embodiments provided by the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the above-described embodiments are merely illustrative. For example, the division of units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in an electrical, mechanical, or other form.

[0129] The unit described as a separation component may or may not be physically separated. The component shown as a unit may or may not be a physical unit, that is, it may be located in one place or may be distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0130] In addition, each functional unit in various embodiments of the present application can be integrated in a processing unit, can also exist physically separately for each unit, or two or more units can be integrated in one unit.

[0131] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the said claims.

Claims

1. A method for inspecting a beneficiation workshop based on an unmanned aerial vehicle, characterized in that, Specifically, it includes the following steps: In the ore dressing workshop, when the drone collects images of equipment at an inclined angle, according to the collected images and their corresponding attitude information, it is judged whether the images collected by the drone will be affected by oblique light irradiation; When the images collected by the drone are affected by oblique light irradiation, identify the areas with shadow interference characteristics in the images and evenly divide them into several sub-areas; Extract the image disturbance characteristic information of each sub-area from the images collected by the drone, and analyze it to determine the shadow gray disturbance degree and disturbance type of each sub-area; According to the disturbance types of each sub-area, respectively execute the corresponding crack recognition path direction judgment strategy and perform dynamic regulation; After completing the dynamic regulation of the crack recognition paths of each sub-area, perform fusion processing on the recognition results, output the complete crack path and its corresponding disturbance level label, and generate an inspection task review list according to the disturbance levels of each sub-area and write it into the scheduling list for subsequent reshooting and path optimization.

2. The ore dressing plant inspection method based on an unmanned aerial vehicle according to claim 1, wherein When the images collected by the drone are affected by oblique light irradiation, identify the areas with shadow interference characteristics in the images. Specifically, when the images collected by the drone are affected by oblique light irradiation, by analyzing the gray values of the collected images, detect the brightness changes in all areas of the images, and combine the distribution of the reflected light sources in the images to identify the areas with shadow interference characteristics in the images. The shadow interference characteristic areas include areas with blurred boundaries, brightness changes, and gray change areas consistent with the light source direction; And evenly divide it into several sub-areas. Specifically, according to the spatial position and size of the shadow interference characteristic areas in the images, use the grid algorithm to evenly divide the shadow interference areas into multiple sub-areas. The grid algorithm dynamically adjusts the division of the sub-areas by evenly distributing grid points in the shadow areas according to the actual area and shape of the shadow areas.

3. The method for inspecting a mineral processing workshop based on a drone according to claim 2, wherein Extract the image disturbance characteristic information of each sub-area from the images collected by the drone, and analyze it to determine the shadow gray disturbance degree and disturbance type of each sub-area. Specifically, it includes the following steps: Extract the image disturbance characteristic information of each sub-area from the images collected by the drone and perform preprocessing after extraction; Extract the gray change characteristic information and edge blur characteristic information from the preprocessed image disturbance characteristic information of each sub-area, and analyze them after extraction to generate the shadow disturbance intensity coefficient and edge blur degree index of each sub-area respectively; Based on the generated shadow disturbance intensity coefficients and edge blur degree indices of each sub-area, construct a shadow gray disturbance evaluation model and generate the shadow gray disturbance coefficient of each sub-area by weighted summation; Determine the preset shadow gray disturbance coefficient threshold interval, and compare it with the generated shadow gray disturbance coefficients of each sub-area after determination. Evaluate the shadow gray disturbance degree of each sub-area according to the comparison result, and determine the disturbance type of each sub-area according to the evaluation result.

4. The method for inspecting a mineral processing workshop based on a drone according to claim 3, characterized in that, The acquisition logic of the shadow disturbance intensity coefficient of each sub-area is as follows: Extract the gray-scale change feature information from the image perturbation feature information of each preprocessed sub-region, specifically including the gray-scale values of each pixel point in each sub-region of the image collected by the drone and the average gray-scale values of all regions adjacent to each sub-region, and calibrate them respectively as and GN i , represents the gray-scale value of the m-th pixel point in the i-th sub-region of the image collected by the drone, and GN i represents the average gray-scale value of all regions adjacent to the i-th sub-region in the image collected by the drone, where i = 1, 2, 3, …, k, m = 1, 2, 3, …, j, and both k and j are positive integers; Calculate the average gray value GPM of all pixel points in each sub-region of the image collected by the drone i , according to the formula: Calculate the difference between the gray value of each pixel point in each sub-region of the image collected by the drone and the average value of the gray values of all pixel points in that sub-region According to the formula: Determine the optimal values of the first adjustment factor γ and the second adjustment factor δ from the actual image data through a data fitting algorithm. The first adjustment factor γ is used to control the response sensitivity of the gray-scale fluctuation within the sub-region to the perturbation coefficient, and the second adjustment factor δ is used to adjust the response amplitude of the gray-scale difference between the sub-region and its neighborhood. Calculate the shadow perturbation intensity coefficient of each sub-region. The specific calculation formula is as follows: where SDIC i is the shadow perturbation intensity coefficient of the i-th sub-region.

5. The method for inspecting a mineral processing workshop based on an unmanned aerial vehicle according to claim 4, wherein, The acquisition logic of the edge blur index of each sub-region is as follows: Extract edge blur feature information from the image perturbation feature information of each preprocessed sub-region, specifically including the edge gradient value and the gray value of each edge pixel point in each sub-region of the image collected by the drone, and calibrate them respectively as and represents the edge gradient value of the nth edge pixel point in the ith sub-region of the image collected by the drone, represents the gray value of the nth edge pixel point in the ith sub-region of the image collected by the drone, where i = 1, 2, 3, …, k, n = 1, 2, 3, …, h, and both k and h are positive integers; Calculate the mean gray value BGP of all edge pixel points in each sub-region of the image collected by the UAV i , according to the formula: Calculate the difference between the gray value of each edge pixel point in each sub-region of the image collected by the drone and the average value of the gray values of all edge pixel points in that sub-region According to the formula: Determine the optimal value of the exponential adjustment factor α from the actual image data through a data fitting algorithm. The adjustment factor α is used to control the response enhancement degree of the edge gradient to the edge blur index. Calculate the edge blur index of each sub-region. The specific calculation formula is as follows: where, EBI i is the edge blur index of the i-th sub-region.

6. The method for inspecting a beneficiation workshop based on an unmanned aerial vehicle according to claim 5, wherein Based on the shadow disturbance intensity coefficient (SDIC) of each generated sub-region i and the edge blurriness index (EBI) i , a shadow gray-scale disturbance evaluation model is constructed, and the shadow gray-scale disturbance coefficient of each sub-region is generated by weighted summation. The specific calculation formula is as follows: SGDC i = ω1 * SDIC i + ω2 * EBI i wherein, SGDC i is the shadow gray-scale perturbation coefficient of the i-th sub-region, and ω1 and ω2 are the shadow perturbation intensity coefficients SDIC i and the edge blur index EBI i of non-zero weight coefficients, and ω1 + ω2 = 1.

7. The ore dressing workshop inspection method based on an unmanned aerial vehicle according to claim 6, wherein, Determine the pre-set threshold interval of the shadow gray-scale disturbance coefficient [SGDC min , SGDC max , and after determination, compare it with the shadow gray-scale disturbance coefficient SGDC i of each generated sub-region. Evaluate the shadow gray-scale disturbance degree of each sub-region according to the comparison result, and determine the disturbance type of each sub-region according to the evaluation result. The specific comparison and analysis are as follows: If SGDC i <SGDC min , the degree of shadow gray perturbation in this sub-region is low, and the perturbation type of this sub-region is a weak perturbation region; If SGDC min ≤SGDC i ≤SGDC max , the degree of shading gray perturbation of this sub-region is medium perturbation degree, and the perturbation type of this sub-region is medium perturbation region; If SGDC i >SGDC max , the degree of shadow gray-scale perturbation in this sub-region is a high perturbation degree, and the perturbation type of this sub-region is a strong perturbation region.

8. The ore dressing workshop inspection method based on an unmanned aerial vehicle according to claim 7, characterized in that, According to the perturbation type of each sub-region, execute the corresponding crack recognition path orientation judgment strategy, specifically including: For the sub-region with the perturbation type of weak perturbation area, the crack recognition path orientation judgment strategy to be executed is: adopt the default reference path recognition algorithm, directly perform linear fitting based on the gray-scale mutation edge in the image, extract the main crack direction, and extend the crack boundary according to the shortest path principle, without performing redundant filtering, angle correction, and auxiliary image fusion processing. For the sub-region with the perturbation type of medium perturbation area, the crack recognition path orientation judgment strategy to be executed is: on the basis of executing the default path recognition algorithm, superimpose the edge fusion calculation based on multi-angle image acquisition, perform curvature compensation and node smoothing on the discontinuous breakpoints and direction mutations that appear in the path extraction, and adjust the path through the historical recognition trajectory. For the sub-region with the perturbation type of strong perturbation area, the crack recognition path orientation judgment strategy to be executed is: preferentially call the anti-interference enhanced recognition mode, perform background enhancement, contrast adjustment, and artifact suppression processing on the image, and apply the high-order edge reconstruction algorithm based on model learning in the path judgment process. Combine the recognition results of adjacent non-perturbed areas to perform fitting backtracking and anomaly elimination on the recognition path. The specific method of dynamic regulation is: continuously monitor the perturbation type of each sub-region during the generation of the recognition path, switch the corresponding path judgment strategy according to the perturbation type, and introduce a perturbation factor to participate in the path weight calculation during the path generation process to adjust the path orientation priority in real time. If the recognition confidence of a certain sub-region is lower than the set threshold, mark this path segment as a "path to be rechecked" and synchronously write it into the reshooting scheduling list.

Images