Automatic detection method and system based on hidden crack characteristics of photovoltaic module

Abstract

The invention discloses an automatic detection method and system based on the hidden crack characteristic of a photovoltaic module, and belongs to the technical field of automatic detection.The method comprises the steps that an unmanned aerial vehicle carries an infrared thermal imager, a polarization camera and a vibration sensor, heat distribution, surface texture and micro-vibration signals of the photovoltaic module are synchronously collected, and a multi-dimensional hidden crack characteristic data set is constructed; according to the method, the lightweight convolutional network is deployed on the unmanned aerial vehicle, local data processing and subfissure probability value output are realized, cloud dependence is reduced, delay is reduced, the subfissure probability value is weighted and calculated through the infrared thermal anomaly factor, the polarization texture anomaly factor and the edge sharpness factor, environmental noise is effectively suppressed, the detection accuracy is improved, and the detection efficiency is improved. A multi-angle reinspection task is inserted into a high-probability subfissure area, a low-risk area is quickly scanned, a medium-probability area is decided according to electric quantity and priority, when the number of reinspection areas is large, a global optimal path is recalculated, when high-probability subfissure is detected, the current path is interrupted immediately, the reinspection task is inserted, and delay is avoided.

Description

Technical Field

[0001] The present invention belongs to the field of automated detection technology, and specifically refers to an automated detection method and system based on the characteristics of hidden cracks in photovoltaic modules. Background Art

[0002] As the global energy structure shifts towards low-carbon development, photovoltaic power generation, as one of the core technologies of clean energy, and the long-term stable operation of its components are crucial to the efficiency and economy of power supply;

[0003] However, the existing automated detection of hidden crack characteristics of photovoltaic modules still has certain defects. The existing ones rely on cloud processing, which has problems such as high latency and large bandwidth usage. The drone inspection adopts a fixed height and focal length, which makes it difficult to achieve both high precision and wide coverage. The inspection path is a preset route and cannot be dynamically adjusted according to real-time detection results. The detection only outputs the location of the hidden crack and lacks classification, resulting in inaccurate maintenance strategies. The detection relies on a single scan and has a high misjudgment rate. Therefore, an automated detection method and system based on the hidden crack characteristics of photovoltaic modules are proposed. Summary of the Invention

[0004] The object of the present invention is to provide an automated detection method and system based on the characteristics of hidden cracks in photovoltaic modules, so as to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automated detection method based on the characteristics of hidden cracks in photovoltaic modules, comprising the following steps:

[0006] S1. The drone is equipped with an infrared thermal imager, a polarization camera, and a vibration sensor to simultaneously collect the thermal distribution, surface texture, and micro-vibration signals of the photovoltaic modules to construct a multi-dimensional hidden crack feature dataset;

[0007] S2: Run a lightweight convolutional network on the edge computing node to analyze the thermal anomaly areas in the infrared image and the microcrack texture in the polarization image in real time, and output the hidden crack probability value;

[0008] S3, automatically adjust the UAV flight altitude and camera focal length according to the hidden crack probability value;

[0009] S4. Dynamically update the inspection route based on the detection results, conduct multi-angle re-inspections on high-probability hidden crack areas, and use a fast scanning mode for low-risk areas;

[0010] S5. Fuse vibration sensor data with image analysis results to distinguish electrical cracks from mechanical cracks and generate a classification diagnosis report;

[0011] S6. Perform a local high-precision rescan of the area marked as hidden cracks to verify the detection results and correct the misjudged samples;

[0012] S7. Upload the verified hidden crack samples and environmental parameters to the cloud and feed them back to the edge computing node to update the convolutional network.

[0013] In the S1, the drone is equipped with an infrared thermal imager, a polarization camera and a vibration sensor to synchronously collect the thermal distribution, surface texture and micro-vibration signals of the photovoltaic module to construct a multi-dimensional hidden crack feature data set; a timing GPS module is integrated on the drone to output a second pulse signal as a global time reference, and the PPS signal is distributed to the infrared thermal imager, polarization camera and vibration sensor through a time synchronization controller, triggering the three to collect data simultaneously, and automatically clearing the time counter, and combining with the inertial navigation board to record the real-time position, attitude and speed of the drone; the installation position, angle and mutual relationship of the infrared thermal imager, polarization camera and vibration sensor are pre-calibrated to establish a unified three-dimensional coordinate system, and the position and attitude data of the drone are obtained in real time through the POS system. Combined with the sensor installation parameters, the three-dimensional spatial coordinates of the data collected by each sensor are calculated. The drone hovers above the photovoltaic module. After the GPS triggers the PPS signal, the infrared thermal imager, polarization camera and vibration sensor start collecting data simultaneously, and the collected thermal distribution, polarization texture and vibration signal data are aligned according to the timestamp to construct a multi-dimensional hidden crack feature data set.

[0014] Said S2 runs a lightweight convolutional network through the edge computing node to analyze the thermal anomaly area in the infrared image and the polarization microcrack texture in the polarization image in real time, and outputs the hidden crack probability value; deploys a lightweight edge computing unit on the UAV to perform preliminary processing on the multi-dimensional hidden crack feature data set, calculates the hidden crack probability value based on the infrared thermal anomaly and the polarization microcrack texture, and sets the infrared thermal anomaly factor as T df , the polarization microcrack texture anomaly factor is P ay , the edge sharpness factor is S ed , T df Expressed as T lcl Indicates the average temperature of the target area, T bd Represents the average temperature of the background area on the component surface, P ay Expressed as θ∈{0°, 45°, 90°, 135°} represents the four polarization directions of the polarization camera, I θ represents the pixel intensity under polarization direction θ, I θ,avg represents the average pixel intensity of the background area under the same polarization direction, S ed Expressed as represents the gradient magnitude of the i-th pixel, It represents the sum of the gradient amplitudes of all pixels in the target area, k represents the gradient amplification factor, which is adjusted according to the texture contrast of the photovoltaic module surface, and the value of k is k∈[1.2,2.0].

[0015] The S2, according to the infrared thermal anomaly factor T df , polarization microcrack texture abnormality factor P ay and edge sharpness factor S ed Calculate the hidden crack probability value, which is realized as:

[0016]

[0017] In the formula, P ck represents the probability value of hidden crack, exp() represents the natural exponential function, α, β, γ represent weight parameters, α takes α∈[0.1,0.5], β takes β∈[0.2,0.8], γ takes γ∈[0.1,0.5], T df represents the infrared anomaly factor, P ay represents the polarization microcrack texture anomaly factor, S ed Represents the edge sharpness factor.

[0018] Among them, the S3 obtains the probability value of hidden cracks and divides the area into low probability area, medium probability area and high probability area according to the probability value. When it is a low probability area, the drone gradually increases the flight altitude to expand the coverage of a single shot. When a high probability hidden crack area is detected, the drone gradually lowers the flight altitude. In the high probability hidden crack area, the camera switches to telephoto mode to focus on the details of the target area. In the low probability area, the camera switches to wide-angle mode. Through GPS, inertial measurement unit and visual feedback, the flight altitude and focal length are corrected in real time. Each frame of the image triggers a hidden crack probability calculation, and the drone state is dynamically adjusted according to the result.

[0019] Among them, the S4 dynamically updates the inspection path based on the detection results, conducts multi-angle re-inspections on high-probability hidden crack areas, and adopts a fast scanning mode for low-risk areas. In high-probability hidden crack areas, the drone automatically triggers the multi-angle re-inspection task after completing the preliminary scan, and generates a new flight path around the target area. In low-probability areas, the drone skips the detailed re-inspection and directly enters the fast scanning mode, quickly covering the preset route. The medium-probability area is determined according to the remaining power and task priority. When a high-probability hidden crack area is detected, the current path is immediately interrupted and a re-inspection task is inserted. A dynamic threshold for the number of re-inspection areas is pre-set. When the number of re-inspection areas is large, the global optimal path is recalculated, and the drone hovers to asynchronously perform global path optimization.

[0020] Among them, the S5 integrates the vibration sensor data and the image analysis results to distinguish between electrical and mechanical cracks, and generates a classification diagnosis report; obtains multi-angle image data and vibration sensor data, extracts the vibration characteristics and image texture characteristics of the electrical and mechanical cracks respectively, matches the characteristics of the vibration data with the texture characteristics of the image, analyzes whether there is a coupling mode, the vibration characteristics of the electrical cracks are highly correlated with thermal anomalies, and the vibration characteristics of the mechanical cracks are highly correlated with the crack morphology, quantifies the coupling strength of the vibration characteristics and the image characteristics through the Pearson correlation coefficient, distinguishes the two types of cracks, classifies them according to the analysis and matching, generates a classification diagnosis report, and marks the crack area.

[0021] The second pulse signal is output by the timing GPS module to trigger the infrared thermal imager, polarization camera and vibration sensor to collect data synchronously. The time stamp accuracy is 1ms. The real-time attitude of the drone is obtained through the POS system. The three-dimensional space coordinates are calculated based on the sensor installation parameters to ensure that the data is aligned in the spatial dimension. The micro-vibration peak point x in the vibration signal is set k ,y k , let a, b be the fitting model parameters, t be the threshold, and the formula for determining the inlier distance between vibration data and image data is:

[0022]

[0023] In the formula, x k ,y k Indicates the micro-vibration peak point in the vibration signal, a, b are fitting model parameters, t is the threshold, when d ck ≤t, it is judged as an interior point;

[0024] According to d ck The effective distance of screening, aligning the time axis of the vibration signal and the image sequence, eliminating nonlinear misalignment, and the path of dynamic time warping are implemented as follows:

[0025] D(n,m)=d(T(n),R(m)+min{D(n-1,m),D(n-1,m-1),D(n,m-1)}),

[0026] In the formula, D(n,m) represents the dynamic regularization path, d(T(n),R(m)) represents the Euclidean distance, and the difference between the nth frame of the vibration signal T(n) and the nth frame of the image sequence R(m) is calculated. min{} represents removing the minimum value to ensure the optimal path. Let the vibration feature vector be V and the image feature vector be I. Calculate the projection ratio similarity between the vibration feature vector V and the image feature vector I. The implementation formula is:

[0027]

[0028] In the formula, Sim(V,I) represents the projection ratio similarity between the dynamic feature vector V and the image feature vector I, and n represents the feature dimension.

[0029] Among them, the S6 performs local high-precision rescanning on the area marked as hidden cracks to verify the detection results and correct the misjudgment samples; obtains the marked hidden crack area, and performs local high-precision rescanning based on the hidden crack probability value being between the preset thresholds. When the matching degree between the classification result and the sensor data is lower than the preset threshold, the manual review process is triggered, and the misjudgment results are verified and corrected.

[0030] Among them, the S7 uploads the verified hidden crack samples and environmental parameters to the cloud, obtains the verified hidden crack samples, and uploads them to the remote end through the HTTP protocol; aggregates the local model updates of multiple drones through horizontal federated learning, and the cloud serves as the receiver of the model gradients of each edge node. After the drone is updated, it is tested with known hidden crack samples to compare the accuracy of the new and old models.

[0031] Among them, an automated detection system based on the hidden crack characteristics of photovoltaic modules is characterized by: including a data acquisition module, an edge computing and real-time analysis module, a flight status adjustment module, a dynamic path planning module, a classification diagnosis module, a result verification and correction module and a cloud-based collaborative optimization module.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. This invention deploys a lightweight convolutional network on the drone to achieve local data processing and output of hidden crack probability values, reducing cloud dependence and latency. It also calculates the hidden crack probability value by weighting the infrared thermal anomaly factor, polarization texture anomaly factor, and edge sharpness factor, effectively suppressing environmental noise and improving detection accuracy.

[0034] 2. This invention balances coverage and resolution by dynamically adjusting flight altitude and camera focal length based on the probability of hidden cracks. This adjustment is triggered once per frame through GPS, IMU, and visual feedback, ensuring dynamic response. This increases coverage in low-risk areas, focuses on details in high-risk areas, and reduces redundant scanning.

[0035] 3. This invention inserts multi-angle re-inspection tasks into high-probability hidden crack areas, quickly scans low-risk areas, and makes decisions based on power and priority in medium-probability areas. When there are many re-inspection areas, the global optimal path is recalculated to ensure efficiency. When a high-probability hidden crack is detected, the current path is immediately interrupted and a re-inspection task is inserted to avoid delays.

[0036] 4. The present invention verifies the detection results by performing local high-precision rescanning of the hidden crack area. When the classification results are significantly inconsistent with the sensor data, a manual review process is triggered to correct the misjudged samples and update the local database to improve the accuracy of subsequent detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is the operating process of an automated detection method based on the characteristics of photovoltaic module cracks in the present invention Figure 1 ;

[0038] Figure 2 This is the operating process of an automated detection method based on the characteristics of photovoltaic module cracks in the present invention Figure 2 ;

[0039] Figure 3 This is the operating process of an automated detection method based on the characteristics of photovoltaic module cracks in the present invention Figure 3 ;

[0040] Figure 4 This is the operating process of an automated detection method based on the characteristics of photovoltaic module cracks in the present invention Figure 4 ;

[0041] Figure 5 The figure is a structural diagram of an automatic detection system based on the hidden crack characteristics of photovoltaic modules according to the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] Example

[0044] See also Figure 1-Figure 5 As shown, the present invention provides a technical solution: comprising the following steps:

[0045] S1. The drone is equipped with an infrared thermal imager, a polarization camera, and a vibration sensor to simultaneously collect the thermal distribution, surface texture, and micro-vibration signals of the photovoltaic modules to construct a multi-dimensional hidden crack feature dataset;

[0046] S2: Run a lightweight convolutional network on the edge computing node to analyze the thermal anomaly areas in the infrared image and the microcrack texture in the polarization image in real time, and output the hidden crack probability value;

[0047] S3, automatically adjust the UAV flight altitude and camera focal length according to the hidden crack probability value;

[0048] S4. Dynamically update the inspection route based on the detection results, conduct multi-angle re-inspections on high-probability hidden crack areas, and use a fast scanning mode for low-risk areas;

[0049] S5. Fuse vibration sensor data with image analysis results to distinguish electrical cracks from mechanical cracks and generate a classification diagnosis report;

[0050] S6. Perform a local high-precision rescan of the area marked as hidden cracks to verify the detection results and correct the misjudged samples;

[0051] S7. Upload the verified hidden crack samples and environmental parameters to the cloud and feed them back to the edge computing node to update the convolutional network.

[0052] In the S1, the drone is equipped with an infrared thermal imager, a polarization camera and a vibration sensor to synchronously collect the thermal distribution, surface texture and micro-vibration signals of the photovoltaic module to construct a multi-dimensional hidden crack feature data set; a timing GPS module is integrated on the drone to output a second pulse signal as a global time reference, and the PPS signal is distributed to the infrared thermal imager, polarization camera and vibration sensor through a time synchronization controller, triggering the three to collect data simultaneously, and automatically clearing the time counter, and combining with the inertial navigation board to record the real-time position, attitude and speed of the drone; the installation position, angle and mutual relationship of the infrared thermal imager, polarization camera and vibration sensor are pre-calibrated to establish a unified three-dimensional coordinate system, and the position and attitude data of the drone are obtained in real time through the POS system. Combined with the sensor installation parameters, the three-dimensional spatial coordinates of the data collected by each sensor are calculated. The drone hovers above the photovoltaic module. After the GPS triggers the PPS signal, the infrared thermal imager, polarization camera and vibration sensor start collecting data simultaneously, and the collected thermal distribution, polarization texture and vibration signal data are aligned according to the timestamp to construct a multi-dimensional hidden crack feature data set.

[0053] Said S2 runs a lightweight convolutional network through the edge computing node to analyze the thermal anomaly area in the infrared image and the polarization microcrack texture in the polarization image in real time, and outputs the hidden crack probability value; deploys a lightweight edge computing unit on the UAV to perform preliminary processing on the multi-dimensional hidden crack feature data set, calculates the hidden crack probability value based on the infrared thermal anomaly and the polarization microcrack texture, and sets the infrared thermal anomaly factor as T df , the polarization microcrack texture anomaly factor is P ay , the edge sharpness factor is S ed , T df Expressed as T lcl Indicates the average temperature of the target area, T bd Represents the average temperature of the background area on the component surface, P ay Expressed as θ∈{0°, 45°, 90°, 135°} represents the four polarization directions of the polarization camera, I θ represents the pixel intensity under polarization direction θ, I θ,avg represents the average pixel intensity of the background area under the same polarization direction, S ed Expressed as represents the gradient magnitude of the i-th pixel, It represents the sum of the gradient amplitudes of all pixels in the target area, k represents the gradient amplification factor, which is adjusted according to the texture contrast of the photovoltaic module surface, and the value of k is k∈[1.2,2.0].

[0054] The S2, according to the infrared thermal anomaly factor T df , polarization microcrack texture abnormality factor P ay and edge sharpness factor S ed Calculate the hidden crack probability value, which is realized as:

[0055]

[0056] In the formula, P ck represents the probability value of hidden crack, exp() represents the natural exponential function, α, β, γ represent weight parameters, α takes α∈[0.1,0.5], β takes β∈[0.2,0.8], γ takes γ∈[0.1,0.5], T df represents the infrared anomaly factor, P ay represents the polarization microcrack texture anomaly factor, S ed Represents the edge sharpness factor.

[0057] Among them, the S3 obtains the probability value of hidden cracks and divides the area into low probability area, medium probability area and high probability area according to the probability value. When it is a low probability area, the drone gradually increases the flight altitude to expand the coverage of a single shot. When a high probability hidden crack area is detected, the drone gradually lowers the flight altitude. In the high probability hidden crack area, the camera switches to telephoto mode to focus on the details of the target area. In the low probability area, the camera switches to wide-angle mode. Through GPS, inertial measurement unit and visual feedback, the flight altitude and focal length are corrected in real time. Each frame of the image triggers a hidden crack probability calculation, and the drone state is dynamically adjusted according to the result.

[0058] Among them, the S4 dynamically updates the inspection path based on the detection results, conducts multi-angle re-inspections on high-probability hidden crack areas, and adopts a fast scanning mode for low-risk areas. In high-probability hidden crack areas, the drone automatically triggers the multi-angle re-inspection task after completing the preliminary scan, and generates a new flight path around the target area. In low-probability areas, the drone skips the detailed re-inspection and directly enters the fast scanning mode, quickly covering the preset route. The medium-probability area is determined according to the remaining power and task priority. When a high-probability hidden crack area is detected, the current path is immediately interrupted and a re-inspection task is inserted. A dynamic threshold for the number of re-inspection areas is pre-set. When the number of re-inspection areas is large, the global optimal path is recalculated, and the drone hovers to asynchronously perform global path optimization.

[0059] Among them, the S5 integrates the vibration sensor data and the image analysis results to distinguish between electrical and mechanical cracks, and generates a classification diagnosis report; obtains multi-angle image data and vibration sensor data, extracts the vibration characteristics and image texture characteristics of the electrical and mechanical cracks respectively, matches the characteristics of the vibration data with the texture characteristics of the image, analyzes whether there is a coupling mode, the vibration characteristics of the electrical cracks are highly correlated with thermal anomalies, and the vibration characteristics of the mechanical cracks are highly correlated with the crack morphology, quantifies the coupling strength of the vibration characteristics and the image characteristics through the Pearson correlation coefficient, distinguishes the two types of cracks, classifies them according to the analysis and matching, generates a classification diagnosis report, and marks the crack area.

[0060] The second pulse signal is output by the timing GPS module to trigger the infrared thermal imager, polarization camera and vibration sensor to collect data synchronously. The time stamp accuracy is 1ms. The real-time attitude of the drone is obtained through the POS system. The three-dimensional space coordinates are calculated based on the sensor installation parameters to ensure that the data is aligned in the spatial dimension. The micro-vibration peak point x in the vibration signal is set k ,y k , let a, b be the fitting model parameters, t be the threshold, and the formula for determining the inlier distance between vibration data and image data is:

[0061]

[0062] In the formula, x k ,y k Indicates the micro-vibration peak point in the vibration signal, a, b are fitting model parameters, t is the threshold, when d ck ≤t, it is judged as an interior point;

[0063] According to d ck The effective distance of screening, aligning the time axis of the vibration signal and the image sequence, eliminating nonlinear misalignment, and the path of dynamic time warping are implemented as follows:

[0064] D(n,m)=d(T(n),R(m)+min{D(n-1,m),D(n-1,m-1),D(n,m-1)}),

[0065] In the formula, D(n,m) represents the dynamic regularization path, d(T(n),R(m)) represents the Euclidean distance, and the difference between the nth frame of the vibration signal T(n) and the nth frame of the image sequence R(m) is calculated. min{} represents removing the minimum value to ensure the optimal path. Let the vibration feature vector be V and the image feature vector be I. Calculate the projection ratio similarity between the vibration feature vector V and the image feature vector I. The implementation formula is:

[0066]

[0067] In the formula, Sim(V,I) represents the projection ratio similarity between the dynamic feature vector V and the image feature vector I, and n represents the feature dimension.

[0068] Among them, the S6 performs local high-precision rescanning on the area marked as hidden cracks to verify the detection results and correct the misjudgment samples; obtains the marked hidden crack area, and performs local high-precision rescanning based on the hidden crack probability value being between the preset thresholds. When the matching degree between the classification result and the sensor data is lower than the preset threshold, the manual review process is triggered, and the misjudgment results are verified and corrected.

[0069] Among them, the S7 uploads the verified hidden crack samples and environmental parameters to the cloud, obtains the verified hidden crack samples, and uploads them to the remote end through the HTTP protocol; aggregates the local model updates of multiple drones through horizontal federated learning, and the cloud serves as the receiver of the model gradients of each edge node. After the drone is updated, it is tested with known hidden crack samples to compare the accuracy of the new and old models.

[0070] Among them, an automated detection system based on the characteristics of hidden cracks in photovoltaic modules is characterized by: including a data acquisition module, an edge computing and real-time analysis module, a flight state adjustment module, a dynamic path planning module, a classification diagnosis module, a result verification and correction module, and a cloud-based collaborative optimization module;

[0071] The cloud-based collaborative optimization module drone uploads hidden crack samples and environmental parameters to the cloud via the HTTPS protocol. The cloud uses the FedAvg algorithm to aggregate the model gradients of multiple drones and update the global convolutional network parameters. The cloud sends the updated model difference package to each drone to reduce bandwidth usage. The drone uses known hidden crack samples to test the new and old models and calculate the accuracy improvement. If the accuracy improvement is greater than 2%, the new model is activated; otherwise, the original model is retained.

[0072] Working Principle: The system uses a drone equipped with an infrared thermal imager, a polarization camera, and a vibration sensor. The system uses a timing GPS module to output a pulse-per-second signal, triggering the three sensors to collect data synchronously. Combined with the inertial navigation board and sensor calibration parameters, the collected thermal, optical, and vibration data are mapped into a unified three-dimensional coordinate system to construct a multi-dimensional hidden crack feature dataset containing spatial location information.

[0073] By running a lightweight convolutional network model on the edge computing node of the drone, multi-dimensional data is processed in real time. By comparing the average temperature difference between the target area and the background area, thermal distribution anomalies are identified, the difference between the pixel intensity and the background in the polarization image is analyzed, and the texture characteristics of microcracks are captured. The clarity of the crack edge is quantified by combining the sum of the image gradient amplitude and the magnification factor. The thermal anomaly factor, texture anomaly factor and edge sharpness factor are integrated, and the hidden crack probability value is output through weighted calculation. The altitude and focal length are adaptively adjusted according to the hidden crack probability value. In low-probability areas, the flight altitude is increased, the camera is switched to wide-angle mode, and a large area is quickly scanned. In high-probability areas, the flight altitude is lowered, the camera is switched to telephoto mode, and the details are focused. Through GPS, inertial measurement unit and visual feedback, a hidden crack probability calculation is triggered for each frame of the image, and the drone status is dynamically adjusted. According to the detection results The inspection path is updated dynamically. When the detection result is a high-probability hidden crack area, a multi-angle re-inspection task is inserted to generate a local spiral grid path. When the detection result is a low-probability area, the detailed re-inspection is skipped and the coverage is quickly covered along the preset route. When the detection result is a medium-probability area, whether to re-inspect is determined based on the remaining power and task priority. When the number of re-inspected areas is large, the global optimal path is recalculated to balance coverage and energy consumption. Through the coupling analysis of vibration data and image features, a classification diagnosis report is generated, and the hidden crack area is marked. The area marked as hidden crack is scanned locally with high precision. When the classification result is significantly different from the sensor data, the manual review process is triggered, the misjudged samples are corrected and the local database is updated. The verified hidden crack samples and environmental parameters are uploaded to the cloud through the HTTP protocol. After the drone is updated, the accuracy of the new and old models is tested using known hidden crack samples.

[0074] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

[0075] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An automated detection method based on the characteristics of hidden cracks in photovoltaic modules, characterized in that: The following steps are involved: S1. The drone is equipped with an infrared thermal imager, a polarization camera, and a vibration sensor to simultaneously collect the thermal distribution, surface texture, and micro-vibration signals of the photovoltaic modules to construct a multi-dimensional hidden crack feature dataset; S2: Run a lightweight convolutional network on the edge computing node to analyze the thermal anomaly areas in the infrared image and the microcrack texture in the polarization image in real time, and output the hidden crack probability value; S3, automatically adjust the UAV flight altitude and camera focal length according to the hidden crack probability value; S4. Dynamically update the inspection route based on the detection results, conduct multi-angle re-inspections on high-probability hidden crack areas, and use a fast scanning mode for low-risk areas; S5. Fuse vibration sensor data with image analysis results to distinguish electrical cracks from mechanical cracks and generate a classification diagnosis report; S6. Perform a local high-precision rescan of the area marked as hidden cracks to verify the detection results and correct the misjudged samples; S7. Upload the verified hidden crack samples and environmental parameters to the cloud and feed them back to the edge computing node to update the convolutional network.

2. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: In the S1, the drone is equipped with an infrared thermal imager, a polarization camera, and a vibration sensor to synchronously collect the thermal distribution, surface texture, and micro-vibration signals of the photovoltaic modules to construct a multi-dimensional hidden crack feature dataset; a timing GPS module is integrated on the drone to output a pulse per second signal as a global time reference. The PPS signal is distributed to the infrared thermal imager, polarization camera, and vibration sensor through a time synchronization controller, triggering the three to simultaneously collect data and automatically clearing the time counter. In combination with the inertial navigation board, the real-time position, attitude, and speed of the drone are recorded; The installation positions, angles, and relationships of the infrared thermal imager, polarization camera, and vibration sensor were pre-calibrated to establish a unified three-dimensional coordinate system. The position and attitude data of the drone were acquired in real time through the POS system. Combined with the sensor installation parameters, the three-dimensional spatial coordinates of the data collected by each sensor were calculated. The drone hovered above the photovoltaic module. After the GPS triggered the PPS signal, the infrared thermal imager, polarization camera, and vibration sensor began to collect data simultaneously. The collected thermal distribution, polarization texture, and vibration signal data were aligned according to the timestamps to construct a multi-dimensional hidden crack feature dataset.

3. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: Said S2 runs a lightweight convolutional network through the edge computing node to analyze the thermal anomaly area in the infrared image and the polarization microcrack texture in the polarization image in real time, and outputs the hidden crack probability value; deploys a lightweight edge computing unit on the UAV to perform preliminary processing on the multi-dimensional hidden crack feature data set, calculates the hidden crack probability value based on the infrared thermal anomaly and the polarization microcrack texture, and sets the infrared thermal anomaly factor as T df , the polarization microcrack texture anomaly factor is P ay , the edge sharpness factor is S ed , T df Expressed as T lcl Indicates the average temperature of the target area, T bd Represents the average temperature of the background area on the component surface, P ay Expressed as θ∈{0°,45°,90°,135°} represents the four polarization directions of the polarization camera, I θ represents the pixel intensity under polarization direction θ, I θ,avg represents the average pixel intensity of the background area under the same polarization direction, S ed Expressed as ▽I i represents the gradient magnitude of the i-th pixel, It represents the sum of the gradient amplitudes of all pixels in the target area, k represents the gradient amplification factor, which is adjusted according to the texture contrast of the photovoltaic module surface, and the value of k is k∈[1.2,2.0].

4. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 3, characterized in that: The S2, according to the infrared thermal anomaly factor T df , polarization microcrack texture abnormality factor P ay and edge sharpness factor S ed Calculate the hidden crack probability value, which is realized as: In the formula, P ck represents the probability value of hidden crack, exp() represents the natural exponential function, α, β, γ represent weight parameters, α takes α∈[0.1,0.5], β takes β∈[0.2,0.8], γ takes γ∈[0.1,0.5], T df represents the infrared anomaly factor, P ay represents the polarization microcrack texture anomaly factor, S ed Represents the edge sharpness factor.

5. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: The S3 obtains the probability value of hidden cracks and divides the area into low probability area, medium probability area and high probability area according to the probability value. When it is a low probability area, the drone gradually increases the flight altitude to expand the coverage of a single shot. When a high probability hidden crack area is detected, the drone gradually lowers the flight altitude. In the high probability hidden crack area, the camera switches to telephoto mode to focus on the details of the target area. In the low probability area, the camera switches to wide-angle mode. The flight altitude and focal length are corrected in real time through GPS, inertial measurement unit and visual feedback. Each frame of the image triggers a hidden crack probability calculation, and the drone state is dynamically adjusted according to the result.

6. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: The S4 dynamically updates the inspection path based on the detection results, conducts multi-angle re-inspections on high-probability hidden crack areas, and adopts a fast scanning mode for low-risk areas. In high-probability hidden crack areas, the drone automatically triggers the multi-angle re-inspection task after completing the preliminary scan, and generates a new flight path around the target area. In low-probability areas, the drone skips the detailed re-inspection and directly enters the fast scanning mode, quickly covering the preset route. The medium-probability area is determined according to the remaining power and task priority. When a high-probability hidden crack area is detected, the current path is immediately interrupted and a re-inspection task is inserted. A dynamic threshold for the number of re-inspection areas is pre-set. When the number of re-inspection areas is large, the global optimal path is recalculated, and the drone hovers to asynchronously perform global path optimization.

7. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: The S5 integrates the vibration sensor data and the image analysis results to distinguish between electrical and mechanical cracks, and generates a classification diagnosis report; obtains multi-angle image data and vibration sensor data, extracts the vibration characteristics and image texture characteristics of the electrical and mechanical cracks respectively, matches the characteristics of the vibration data with the texture characteristics of the image, analyzes whether there is a coupling mode, finds that the vibration characteristics of the electrical cracks are highly correlated with thermal anomalies, and that the vibration characteristics of the mechanical cracks are highly correlated with the crack morphology, quantifies the coupling strength of the vibration characteristics and the image characteristics through the Pearson correlation coefficient, distinguishes the two types of cracks, classifies them according to the analysis and matching, generates a classification diagnosis report, and marks the crack areas.

8. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: The S6 performs a local high-precision rescan on the area marked as a hidden crack to verify the detection results and correct the misjudgment samples; obtains the marked hidden crack area, and performs a local high-precision rescan based on the hidden crack probability value being between the preset thresholds. When the matching degree between the classification result and the sensor data is lower than the preset threshold, the manual review process is triggered, and the misjudgment results are verified and corrected.

9. The automated detection method based on the characteristics of hidden cracks in photovoltaic modules according to claim 1, characterized in that: The S7 uploads the verified hidden crack samples and environmental parameters to the cloud, obtains the verified hidden crack samples, and uploads them to the remote end via the HTTP protocol; aggregates the local model updates of multiple drones through horizontal federated learning, and the cloud serves as the receiver of the model gradients of each edge node. After the drone is updated, it is tested with known hidden crack samples to compare the accuracy of the new and old models.

10. An automated detection system based on the characteristics of hidden cracks in photovoltaic modules implemented according to the method of claim 1, characterized in that: It includes data acquisition module, edge computing and real-time analysis module, flight status adjustment module, dynamic path planning module, classification diagnosis module, result verification and correction module and cloud-based collaborative optimization module.

Images