Self-adaptive cleaning control method of cleaning robot based on image texture analysis

Abstract

The present application belongs to the technical field of image analysis, and particularly relates to a self-adaptive cleaning control method for a cleaning robot based on image texture analysis, which comprises the following steps: performing top-hat transformation on an original image by using a multi-angle linear structural element to obtain an anisotropic residual texture image and calculate a gray run matrix; constructing a gray run inverse coupling inertia based on a negative correlation mapping relationship between gray and run; obtaining a morphological erosion residual ratio representing stain stability through a morphological erosion operation, and then fusing to generate a sediment consolidation modulus; and adjusting the command marching speed and the command shear torque of the robot accordingly. The present application solves the problems of grid line interference and misjudgment caused by reflection by deeply coupling visual features and physical stability, and realizes accurate matching between cleaning intensity and stain state.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology. More specifically, this invention relates to an adaptive cleaning control method for cleaning robots based on image texture analysis. Background Technology

[0002] As a crucial component of renewable energy, the cleanliness of photovoltaic (PV) panel surfaces directly impacts the energy conversion efficiency of PV modules. In outdoor operating environments, PV panel surfaces easily accumulate various types of deposits and stains. These stains include not only loose dust but also stubborn deposits such as bird droppings, mud stains, or long-dried ash. To maintain power generation efficiency, the industry widely employs cleaning robots to replace manual labor for automated operations. The core prerequisite for achieving adaptive adjustment of cleaning intensity is the ability to accurately monitor the state of contamination on the panel surface using visual sensors.

[0003] Existing technologies typically utilize image processing to assess the degree of contamination, primarily characterizing stain distribution through indicators such as the long run dominance factor or grayscale nonuniformity in the statistical grayscale run-length matrix. However, these standard statistical indicators have significant physical limitations in complex photovoltaic scenarios. First, photovoltaic panels themselves have a dense background of metal grid lines and dark patterns on the cells, which overlap with the deposited stains in terms of grayscale distribution. This causes standard texture indicators to easily misjudge regular grid line shadows as connected, hardened stains. Second, existing visual indicators often treat color information and geometric connectivity as independent variables, lacking a deep characterization of the physical density of the stains. They cannot identify impurities with similar visual features but vastly different removal difficulties, preventing robots from specifically adjusting the output parameters of mechanical actuators.

[0004] The core technical challenge arising from this is that existing detection methods cannot accurately identify the physical consolidation strength of stains under complex background interference from photovoltaic panels and translate it into effective mechanical control logic. This leads to a technical dilemma for cleaning robots in actual operation: over-cleaning ordinary dust results in energy loss and damage to the solar cells, while insufficient cleaning of stubborn deposits leads to hot spot effects on the panels. Summary of the Invention

[0005] To address the technical problem of existing technologies failing to accurately identify stain consolidation strength and achieve precise mechanical control against complex photovoltaic panel backgrounds, this invention provides an adaptive cleaning control method for a cleaning robot based on image texture analysis. The method includes: acquiring an original image of the photovoltaic panel; processing the original image using linear structuring elements at multiple angles to generate an anisotropic residual texture map stripped of directional background; and calculating a grayscale run-length matrix; based on the negative correlation between run length and grayscale level in the grayscale run-length matrix, calculating the grayscale run-length inverse coupling inertia to extract dark, long connected regions. The corresponding hardened stain characteristics are analyzed; multiple morphological erosion operations are performed on the anisotropic residual texture map using disk structural elements; based on the ratio of the grayscale of the eroded image after erosion to that of the anisotropic residual texture map before erosion, the morphological erosion residual ratio, which characterizes the physical structural stability of the stain, is obtained; the morphological erosion residual ratio is nonlinearly amplified and inversely coupled with the grayscale run inertia to obtain the sediment consolidation modulus used to quantify the stain's resistance to mechanical stripping; based on the sediment consolidation modulus, the command travel speed and command shear torque are calculated to achieve adaptive adjustment of cleaning intensity.

[0006] This invention successfully establishes a correlation between color depth and connectivity length through grayscale run-length inverse coupling inertia. Utilizing the physical characteristic that heavy stains on photovoltaic panels inevitably exhibit low grayscale values ​​due to light obstruction, it solves the problem in existing technologies where long grid lines and shadows are easily confused with solid stains, significantly improving the accuracy of target recognition. Simultaneously, the introduction of morphological erosion residue ratio within the algorithm pre-simulates the stain ablation process under mechanical force, enabling the robot to accurately distinguish impurities with similar visual features but vastly different physical properties, bridging the mapping gap between visual signals and cleaning mechanical force requirements. Furthermore, a nonlinear control strategy based on the sediment consolidation modulus, using a hyperbolic tangent function for smooth velocity adjustment and a logarithmic function for torque-constrained enhancement, ensures both high efficiency in handling ordinary dust areas and the ability to eradicate stains and avoid panel damage when encountering stubborn hardened materials, effectively guaranteeing the safety and reliability of outdoor working environments.

[0007] Preferably, the formula for calculating the grayscale run-length reverse coupling inertia is: In the formula, This represents the inverse coupling inertia of the grayscale run; This represents the total number of gray levels. The maximum possible run length; These are the element values ​​in the grayscale run-length matrix; The quantized grayscale level; This refers to the length of the journey. To prevent zero constant; and This is the coupling regulation index.

[0008] This invention introduces power terms and sensitivity factors to construct inverse coupling inertia, which amplifies the weight of dark, long connected regions, enabling it to sensitively lock onto high-risk, thick stains in visual signals and improve the resolution of indicators.

[0009] Preferably, the process of obtaining the morphological erosion residual ratio includes: summing the gray values ​​of all pixels in the eroded image as the residual pixel sum; summing the gray values ​​of all pixels in the anisotropic residual texture map as the reference pixel sum; and calculating the ratio of the residual pixel sum to the reference pixel sum to obtain the morphological erosion residual ratio.

[0010] This invention defines the erosion residue ratio by summing the pixels, quantifying the stain from the perspective of physical structural stability, and solving the problem that simple visual statistics cannot reflect the stain peeling resistance.

[0011] Preferably, the formula for calculating the consolidation modulus of the sediment is: In the formula, Indicates the modulus of consolidation of sediments; This represents the inverse coupling inertia of the grayscale run; Indicates the morphological erosion residue ratio; These are the normalization coefficients; Residual sensitive factors; This represents an exponential function with the natural constant as its base.

[0012] This invention uses a natural exponential function to nonlinearly amplify the residual ratio, constructing a sediment consolidation modulus. This enables the system to generate a step-like early warning response when facing highly dense and highly consolidated stubborn stains, providing safety redundancy for extreme operating conditions.

[0013] Preferably, the formula for calculating the command travel speed is: In the formula, This refers to the command's travel speed; This is the robot's reference speed; It is the hyperbolic tangent function; This is the speed regulation coefficient; It represents the consolidation modulus of sediments.

[0014] This invention utilizes the saturation characteristics of the hyperbolic tangent function to adjust the travel speed, ensuring that the robot can smoothly decelerate to a low-speed crawling state when it detects heavy stains, avoiding transmission shocks caused by frequent motor jumps, and ensuring the lifespan of the equipment.

[0015] Preferably, the formula for calculating the commanded shear torque is: In the formula, This is the command shear torque; This refers to the no-load torque of the motor. This is the torque gain coefficient; It is the natural logarithm function; This is the torque sensitivity coefficient; It represents the consolidation modulus of sediments.

[0016] This invention adjusts the shear torque using a natural logarithmic function, achieving a limited torque growth logic. This ensures sufficient mechanical force to peel off stubborn clumps while setting an upper limit from a dynamic perspective, effectively preventing damage to the photovoltaic glass caused by torque overload.

[0017] Preferably, generating an anisotropic residual texture map stripped of directional background includes: performing morphological opening operations on the original image using linear structuring elements at multiple angles and calculating a top-hat image, taking the minimum value of multiple top-hat images at each spatial coordinate pixel point, and generating an anisotropic residual texture map stripped of directional background.

[0018] This invention utilizes the physical properties of the grid background in a specific geometric direction by setting a specific set of angles to perform a minimum value operation. This minimizes directional interference, ensures the purity of the data base for subsequent analysis, and improves the robustness of the detection.

[0019] Preferably, the method for obtaining the top-hat image is as follows: subtract the gray value of the corresponding coordinate pixel in the opening operation image at each angle from the gray value of each pixel in the original image to obtain the residual component of each pixel, and the residual components of all pixels constitute the top-hat image at the corresponding angle.

[0020] Preferably, the calculation of the grayscale run-length matrix includes: performing grayscale level equal-interval quantization on the anisotropic residual texture map to obtain a 16-level grayscale map; and statistically summing the run-length according to four preset directions to generate a grayscale run-length matrix.

[0021] Preferably, the plurality of angles includes 0 degrees, 45 degrees, 90 degrees, and 135 degrees.

[0022] The beneficial effects of this invention are as follows:

[0023] This invention establishes a deep mapping logic from visual features to physical consolidation by combining multi-angle top-hat transformation to extract residuals, grayscale run-length inverse coupling, and morphological virtual scraping. It effectively removes directional background interference from metal grid lines through anisotropic residual processing, accurately extracts dark-colored slab features by utilizing the negative correlation coupling between grayscale and connectivity, and introduces corrosion residual ratio to pre-simulate the mechanical stripping process. This solves the problem that existing technologies cannot accurately identify the physical consolidation strength of stains and convert it into control logic, and achieves dynamic equivalence matching between cleaning intensity and stain state. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the adaptive cleaning control method for a cleaning robot based on image texture analysis in this invention;

[0025] Figure 2 This is a schematic representation of the original image of a photovoltaic panel;

[0026] Figure 3 This is a schematic representation of anisotropic residual texture maps;

[0027] Figure 4 This is a schematic diagram showing the grid distribution of the command travel speed;

[0028] Figure 5 This is a schematic diagram showing the grid distribution of the command shear torque. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] This invention discloses an adaptive cleaning control method for a cleaning robot based on image texture analysis, referring to... Figure 1 This includes steps S1 to S3:

[0032] S1: Acquire the original image of the photovoltaic panel, perform multi-angle structuring element top-hat transformation on it to obtain an anisotropic residual texture map, and calculate the grayscale run matrix based on the anisotropic residual texture map.

[0033] It should be noted that because the background of the metal grid lines and the dark patterns of the solar cells overlap with the grayscale distribution of deposited stains, directly extracting stain features easily introduces a large amount of background noise, leading to inaccurate determination of the subsequent deposit consolidation degree. Therefore, this invention utilizes the strong horizontal or vertical orientation of the grid lines in spatial distribution, while deposited stains typically exhibit isotropic random edges, and uses linear structuring elements in multiple directions for residual extraction. This operation can strip away the regular directional background, obtaining pure stain texture information, providing a data base for subsequently constructing high-sensitivity evaluation indicators.

[0034] Specifically, high-definition sensors are used to acquire raw images of photovoltaic panels in real time. Construct an angle set Linear structural elements from various angles Based on the linear structure elements of each angle, the original image is processed. Performing a morphological opening operation yields an opening image. ; Calculate the top cap diagram at each angle The minimum value at each spatial coordinate pixel point is taken from multiple top-hat images to generate an anisotropic residual texture map. .

[0035] Furthermore, for anisotropic residual texture maps Perform grayscale level equal spacing quantization to The grayscale run-length matrix is ​​calculated by statistically analyzing and summing data in four directions. ;in, This is a grayscale run-length matrix; This represents the quantized grayscale level. Indicates the length of the journey.

[0036] It should be noted that when obtaining the anisotropic residual texture map, the minimum value of the top-hat transformation results in multiple directions is taken. Since the metal grid lines are effectively fitted and eliminated under the linear structuring element in a specific direction, while irregular stains retain residuals in all directions, the minimum value operation can suppress the interference of the strongly directional metal grid lines to the greatest extent and retain the isotropic stain features. The gray value distribution of the final anisotropic residual texture map reflects the saliency of the stains after the background is stripped, and is positively correlated with the statistical accuracy of the subsequent gray-level run matrix.

[0037] For example, Figure 2 The image shows the initial visual state of the photovoltaic panel before it is processed, including the background of the metal grid lines on the panel surface, the dark patterns of the cells, and the unevenly distributed deposited dirt. The physical background in the image exhibits strong directional linear features, while the deposited dirt shows an isotropic discrete texture distribution in space. Figure 3 An anisotropic residual texture map is presented, showing the visual effect of the original image after multi-angle linear structuring element opening operation and residual extraction. The image highlights the isotropic feature distribution after stripping the directional background, effectively suppressing the interference of regular metal grid lines and dark patterns of solar cells on the photovoltaic panel. The discrete bright areas in the image correspond to the deposited stain texture with isotropic features on the panel surface, providing clean data for subsequent grayscale run matrix statistics and stain saliency analysis.

[0038] S2: Based on the gray-scale run-length matrix and anisotropic residual texture map, construct the gray-scale run-length inverse coupling inertia and morphological erosion residual ratio.

[0039] It should be noted that existing texture statistics only focus on run length and ignore the physical meaning of grayscale values ​​in photovoltaic scenarios, causing algorithms to be unable to distinguish between light-colored reflective stripes and dark, heavy stains, resulting in unreasonable allocation of cleaning pressure. Therefore, this invention establishes a negative correlation between grayscale and length and introduces morphological erosion residue ratio as a virtual scratching experiment at the algorithm level. Since heavy stains have strong morphological retention under erosion operations, while the edges of floating dust disappear quickly, this operation realizes a logical transformation from visual distribution to the density of physical structure, providing a basis for accurately assessing the physical resistance of stains.

[0040] Specifically, based on the grayscale run-length matrix Calculate the grayscale run-length inverse coupling inertia :

[0041]

[0042] In the formula, This represents the inverse coupling inertia of the grayscale run; This represents the total number of gray levels. The maximum possible run length; These are the element values ​​in the grayscale run-length matrix; The quantized grayscale level; This refers to the length of the journey. To prevent the constant from being zero, the value is set to 0.001; and This is the coupling regulation index.

[0043] Wherein, the coupling adjustment index and These are boundary constraint parameters used to control the sensitivity weights of the spatial connectivity and grayscale characteristics of stains in the system; because photovoltaic panels are greatly affected by light and shadow fluctuations and background grid line interference in outdoor environments, if and If the setting is too small, the system will degenerate into a simple linear statistic, resulting in insufficient differentiation between dark, hardened stains and light-colored background shadows, easily leading to missed detections. If the setting is too large, the power effect will drastically amplify local high-frequency noise in the image, causing the calculated value to overflow and triggering the robot frequently and falsely initiating the heavy-duty cleaning mode, resulting in operational oscillations. Therefore, the setting should be... The range of values ​​is to , The range of values ​​is to In this embodiment, Set as , Set as This is to maintain the stability of numerical calculations while significantly widening the gap in characteristic responses; in other embodiments, implementers can adjust these two parameters according to the actual situation.

[0044] The calculation formula uses the run length Place it in the molecule and take the power, while simultaneously setting the gray level. Placed in the denominator, so that the region appears as a dark color (low gray level). It also has long connectivity (large run length) When the grayscale run-length inverse coupling inertia is... The value of increases non-linearly; by increasing the weight of dark, long connected regions, this variable achieves significant extraction of the characteristics of hardened stains and can effectively distinguish between light dust and heavy scaling.

[0045] Furthermore, a disk-shaped structural element with a radius of 3 pixels is set for the anisotropic residual texture map. Perform continuously A series of morphological erosion operations were performed to obtain erosion images. .

[0046] Furthermore, the morphological erosion residue ratio was calculated. :

[0047]

[0048] in, The ratio of morphological erosion residue; To erode the image at pixel coordinates The grayscale value at that location; For anisotropic residual texture maps in pixel coordinates The grayscale value at that location; This indicates a summation operation on all pixels in the image.

[0049] The calculation formula reflects the area retention rate of the stained area after multiple etching processes, when the remaining pixels and When the size is larger, the morphological erosion residue is greater than The larger the value, the stronger the scratch resistance of the stain at the algorithm level, and the denser the physical structure; this is highly consistent with the difficulty of removing stains in the actual mechanical cleaning process.

[0050] It should be noted that the grayscale run-back coupling inertia and the morphological erosion residue ratio quantify the stubbornness of stains from the perspectives of visual coupling characteristics and structural stability, respectively.

[0051] S3: By fusing grayscale run-length inverse coupling inertia and morphological erosion residue ratio, the consolidation modulus of sediment is derived, and the robot's commanded travel speed and commanded shear torque are calculated accordingly.

[0052] It should be noted that, because a single visual indicator is prone to impulsive misjudgments when faced with sudden changes in local lighting, leading to frequent acceleration and deceleration of the robot motors and damage to the transmission system, this invention establishes a sediment consolidation modulus through nonlinear fusion. Since this modulus integrates color depth, area size, and corrosion resistance, it can accurately reflect the equivalent strength of the stain against mechanical abrasion. Based on this, this invention establishes a control law using a hyperbolic tangent function and a logarithmic function to ensure a deep match between cleaning intensity and stain condition.

[0053] Specifically, based on grayscale run-length inverse coupling inertia Compared with morphological erosion residue The consolidation modulus of the sediment was calculated. Sediment consolidation modulus Satisfying the expression:

[0054]

[0055] In the formula, Indicates the modulus of consolidation of sediments; This represents the inverse coupling inertia of the grayscale run; Indicates the morphological erosion residue ratio; These are the normalization coefficients; Residual sensitive factors; This represents an exponential function with the natural constant as its base.

[0056] Wherein, the normalization coefficient This is a range-matching parameter set to quantize and map high-dimensional image texture features to the adjustable range of the standard control function. Since the grayscale run-length inverse coupling inertia is a globally accumulated value obtained based on pixel grayscale and spatial connectivity length statistics, its magnitude is large. If this coefficient is directly deleted without scaling, the sediment consolidation modulus will far exceed the saturation limit of the independent variable in the subsequent control function. This will cause the robot's travel speed command, when faced with even slight dirt, to instantly drop to zero due to the saturation characteristics of the hyperbolic tangent function, resulting in the system losing its normal operating speed. The ability to identify gradients in the degree of contamination can lead to frequent false shutdowns. Furthermore, large feature values ​​directly involved in calculations can easily cause numerical overflow at the controller's underlying layer, resulting in the failure of the adaptive control law. Therefore, the normalization coefficient is set to a range of 0.00001 to 0.001. In this embodiment, the normalization coefficient is set to 0.0001 to ensure that the sediment consolidation modulus is in a highly sensitive linear adjustment response range when encountering typical stubborn contamination. In other embodiments, the implementer can adjust this parameter according to the sensor's resolution and grayscale quantification level.

[0057] Among them, the residual sensitive factor This is a logical weighting parameter set to adjust the proportion of morphological erosion residue ratio in the exponential fusion weighting. Since the results of the virtual scratching experiment directly reflect the physical density of the stain, if this value is set too small, the exponential characteristic of the sediment consolidation modulus will tend to linearize, leading to a significant reduction in the system's ability to distinguish between stains with vastly different properties, such as bird droppings and coal ash. If it is set too large, the modulus calculation will produce extremely strong nonlinear expansion, causing even slightly dense stains to trigger the full-scale control, thus rendering the system incapable of graded quantitative adjustment. Therefore, a residue sensitivity factor is set. The range of values ​​is to In this embodiment, the residual sensitive factor is set to This is to maintain the smoothness of the control quantity calculation while ensuring accurate differentiation of stain types; in other embodiments, implementers can adjust this parameter according to the actual situation.

[0058] The calculation formula utilizes an exponential function to amplify the morphological erosion residue ratio, so that when the stain simultaneously possesses dark connectivity features and a highly dense structure, the variable... and Simultaneously increasing the sediment consolidation modulus It will produce a very large numerical response, and increase rapidly in a nonlinear manner; sediment consolidation modulus The higher the value, the more severe the contamination of the stains on the photovoltaic panel surface, requiring stronger mechanical intervention. This enables precise quantification of highly contaminated stains, providing data support for decision-making.

[0059] Furthermore, based on the sediment consolidation modulus Calculate the speed of the command. With commanded shear torque .

[0060] Among them, the speed of instruction travel The formula for calculation is:

[0061]

[0062] In the formula, This refers to the command's travel speed; This is the robot's reference speed; It is the hyperbolic tangent function; This is the speed regulation coefficient; It represents the consolidation modulus of sediments.

[0063] Wherein, the speed regulation coefficient This is a dynamic adjustment parameter set to determine the slope of the robot's travel speed in response to changes in the consolidation modulus of the sediment. Since the monorail robot requires sufficient scrubbing time when passing through heavily fouled areas, if this value is set too small, the speed response will exhibit a significant lag effect, causing the robot to be unable to decelerate in time, resulting in incomplete cleaning. If set too large, the speed becomes overly sensitive to minute fluctuations in the modulus, causing the robot to generate violent braking feedback even when encountering non-critical dust, leading to impact loads on the travel mechanism and reduced operational efficiency. Therefore, a speed regulation coefficient is set. The range of values ​​is to In this embodiment, the speed regulation coefficient is set to This ensures that the robot can smoothly decelerate to the target creep speed and effectively filter background noise disturbances; in other embodiments, the implementer can adjust this parameter according to the actual situation.

[0064] This calculation formula utilizes the saturation characteristics of the hyperbolic tangent function, allowing the sediment consolidation modulus to change as the sediment consolidation modulus increases. The increase in the speed of instruction execution The robot descends smoothly and monotonously; this adjustment logic ensures that it can actively slow down when faced with stubborn stains, increasing the time the roller brush operates on a unit area.

[0065] For example, Figure 4 The diagram shows the grid distribution of the robot's travel speed on the photovoltaic panel working surface. The numerical state of each grid cell is nonlinearly adjusted by the consolidation modulus of the sediment, reflecting the speed smoothing adjustment logic based on the saturation characteristics of the hyperbolic tangent function. When dark, long connected regions and hardened stains with high physical structural stability are detected, the corresponding grid signal shows a significant numerical attenuation, causing the robot to enter a low-speed crawling state to increase the effective cleaning time. In areas without significant stain interference, the robot maintains the baseline operating speed.

[0066] Among them, the command shear torque The formula for calculation is:

[0067]

[0068] In the formula, This is the command shear torque; This refers to the no-load torque of the motor. This is the torque gain coefficient; It is the natural logarithm function; This is the torque sensitivity coefficient; It represents the consolidation modulus of sediments.

[0069] Wherein, the torque gain coefficient These are actuator constraint parameters set to adjust the intensity of the reaction force against the physical resistance of the system to stains and to control the increase of the shear torque as a function of the sediment consolidation modulus. They were calibrated experimentally, including: artificially simulating stains of different consolidation levels on the surface of a standard voltammetric module, such as loose dust, damp mud, and dried bird droppings, and recording the actual peeling force requirements for each sample; starting the robot into automatic cleaning mode, inputting the sediment consolidation modulus into the control system in real time, and observing the different... The output current fluctuation of the roller brush motor is measured; the minimum driving current point that causes stubborn stains to peel off is found, which is the critical effective torque value under the consolidation modulus of the sediment; by establishing the mapping relationship between the sediment consolidation modulus and the critical torque, it is ensured that... The value of can overcome the adhesion of the hardened material without generating excessive mechanical impact at the moment of contact.

[0070] Wherein, the torque sensitivity coefficient This is an actuator constraint parameter set to control the gain rate of change in the shear torque of the command relative to the degree of consolidation of the stain. Because there is a difference in magnitude between the resistance to peeling stubborn stains and loose dust, if this value is set too small, the torque increase of the roller brush motor will be too slow, failing to output sufficient shear force when facing hard dirt, resulting in the attachment failing to peel off. If set too large, the torque response will be too fast, easily reaching the motor's power limit, causing the drive circuit to overheat and shut down, or scratching the anti-reflective coating of the photovoltaic panel due to excessive instantaneous output torque. Therefore, setting a torque sensitivity coefficient is crucial. The range of values ​​is to In this embodiment, the torque sensitivity coefficient is set to This ensures that the torque quickly reaches the cleaning threshold within the safety boundary and maintains efficient peeling capability; in other embodiments, implementers can adjust this parameter according to actual conditions.

[0071] The calculation formula achieves a constrained increase in torque through a logarithmic function, ensuring that the torque is limited when the sediment consolidation modulus... When raised, the robot can provide greater shearing force from the roller brush to remove stains, while preventing excessive torque from damaging the photovoltaic glass cover, thus achieving a balance between cleaning efficiency and hardware safety.

[0072] In summary, the sediment consolidation modulus achieves a quantitative mapping from visual characteristics to mechanical control parameters; as its value increases, the commanded travel speed increases. Smooth descent, commanded shear torque It exhibits logarithmic growth, achieving the effect of deep cleaning even for heavily soiled areas, while ensuring a balance between cleaning efficiency and safety.

[0073] For example, Figure 5This is a grid distribution diagram of the command shear torque, which shows the output distribution of the cleaning robot's mechanical actuator under different operating coordinates. The response intensity of the grid in the diagram is generated by mapping the consolidation modulus of the sediment through the natural logarithm function, reflecting the limited growth enhancement logic for stubborn deposits. The numerical peak region in the diagram is highly matched with the spatial coordinates of highly stubborn stains, ensuring that the system provides sufficient mechanical force to peel off the deposits while avoiding damage to the photovoltaic glass cover plate due to torque overload through the upper limit setting at the dynamic level.

Claims

1. An adaptive cleaning control method for a cleaning robot based on image texture analysis, characterized in that, include: The original image of the photovoltaic panel is obtained, and the original image is processed using linear structuring elements at multiple angles to generate an anisotropic residual texture map stripped of the directional background, and the grayscale run matrix is ​​calculated. Based on the negative correlation between run length and gray level in the gray run matrix, the reverse coupling inertia of gray run is calculated to extract the slab stain features corresponding to dark long connected regions. Grayscale run-inverse coupling inertia for: ； This represents the total number of gray levels. The maximum possible run length; These are the element values ​​in the grayscale run-length matrix; The quantized grayscale level; This refers to the length of the journey. To prevent zero constant; and The coupling regulation index; Multiple morphological erosion operations were performed on the anisotropic residual texture map using disk structuring elements. Based on the ratio of the grayscale of the eroded image after erosion to that of the anisotropic residual texture map before erosion, the morphological erosion residual ratio, which characterizes the physical structural stability of the stain, was obtained. The morphological erosion residual ratio was nonlinearly amplified and inversely coupled with grayscale run length inertia fusion to obtain the sediment consolidation modulus for quantifying the mechanical stripping resistance of stains, as follows: ； Indicates the modulus of consolidation of sediments; Indicates the morphological erosion residue ratio; These are the normalization coefficients; Residual sensitive factors; Represents an exponential function with the natural constant as its base; Based on the sediment consolidation modulus, the commanded travel velocity and commanded shear torque are calculated to achieve adaptive adjustment of cleaning intensity; Command travel speed for: ； This is the robot's reference speed; It is the hyperbolic tangent function; This is the speed regulation coefficient; The commanded shear torque is: ； This is the command shear torque; This refers to the no-load torque of the motor. This is the torque gain coefficient; It is the natural logarithm function; This is the torque sensitivity coefficient.

2. The adaptive cleaning control method for a cleaning robot based on image texture analysis according to claim 1, characterized in that, The process of obtaining the morphological erosion residual ratio includes: The sum of the gray values ​​of all pixels in the eroded image is taken as the residual pixel sum; The sum of the gray values ​​of all pixels in the anisotropic residual texture map is used as the baseline pixel sum. The ratio of the residual pixel sum to the reference pixel sum is calculated to obtain the morphological erosion residual ratio.

3. The adaptive cleaning control method for a cleaning robot based on image texture analysis according to claim 1, characterized in that, The generation of anisotropic residual texture maps stripped of directional backgrounds includes: Morphological opening operations are performed on the original image using linear structuring elements at multiple angles, and a top-hat image is calculated. The minimum value of multiple top-hat images at each spatial coordinate pixel point is taken to generate an anisotropic residual texture map stripped of directional background.

4. The adaptive cleaning control method for a cleaning robot based on image texture analysis according to claim 3, characterized in that, The method for obtaining the top hat image is as follows: Subtract the gray value of the corresponding coordinate pixel in the opening operation image at each angle from the gray value of each pixel in the original image to obtain the residual component of each pixel. The residual components of all pixels constitute the top-hat image at the corresponding angle.

5. The adaptive cleaning control method for a cleaning robot based on image texture analysis according to claim 1, characterized in that, The calculation of the grayscale run-length matrix includes: The anisotropic residual texture map is quantized with equal spacing between gray levels to obtain a 16-level gray map. Run lengths are statistically analyzed and accumulated according to four preset directions to generate a grayscale run length matrix.

6. The adaptive cleaning control method for a cleaning robot based on image texture analysis according to claim 1, characterized in that, The multiple angles include 0 degrees, 45 degrees, 90 degrees, and 135 degrees.

Images