Unmanned aerial vehicle aerial photo photovoltaic module detection method and system

Abstract

The invention provides an unmanned aerial vehicle aerial photovoltaic module detection method and system, and relates to the technical field of image processing, and the method comprises the steps: obtaining the hyperspectral image data of a photovoltaic module; performing correction processing on the hyperspectral image data to obtain corrected image data; performing dimension reduction on the corrected image data through an independent component analysis algorithm to obtain an independent component image; sampling is carried out on the independent component image, and a training sample is determined; training the U-Net model through the training sample; inputting the whole independent component image into the trained U-Net model, and outputting a classification result graph; according to the classification result graph, the operation state of the photovoltaic module is evaluated, and a photovoltaic module inspection report is output. According to the invention, through the hyperspectral image and the U-Net model, efficient and accurate detection and automatic inspection of the operation state of the photovoltaic module are realized.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for detecting photovoltaic modules by drone aerial photography. Background Technology

[0002] During operation, photovoltaic power plants are susceptible to various malfunctions in their modules due to manufacturing defects, environmental corrosion, and other factors, such as hot spots, microcracks, and dirt blockage. These malfunctions can significantly reduce power generation efficiency and even pose safety hazards. Therefore, regular and efficient inspections of photovoltaic modules are crucial for ensuring the safe and stable operation of the power plant and maximizing investment returns.

[0003] Currently, traditional inspection methods mainly rely on manual visual inspection or infrared temperature measurement cameras. However, drone-based visible light or infrared imaging inspection technology has been widely adopted. Operators use drones to collect visible light or thermal images of the photovoltaic array, which are then observed and analyzed by professionals to identify abnormal areas.

[0004] However, this existing technology has obvious limitations. Visible light images are difficult to detect defects such as hot spots and microcracks in the early stages or inside the components. While conventional infrared images are sensitive to hot spots, they are easily affected by ambient temperature, lighting conditions and surface reflection of the components, resulting in high false detection and false negative rates. Furthermore, they are highly dependent on the experience of the analysts, have low automation, and it is difficult to balance inspection efficiency and accuracy. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method for inspecting photovoltaic modules by drone aerial photography, which can solve the obvious limitations of the prior art. Visible light images are difficult to detect defects such as hot spots and microcracks in the early stage or inside the module. Although conventional infrared images are sensitive to hot spots, they are easily affected by ambient temperature, lighting conditions and surface reflection of the module, resulting in high false detection and false negative rates. In addition, they are highly dependent on the experience of the analysts, have low automation, and it is difficult to balance inspection efficiency and accuracy.

[0006] A first aspect of this invention provides a method for detecting photovoltaic modules using drone aerial photography, comprising:

[0007] S1: Acquire hyperspectral image data of photovoltaic modules;

[0008] S2: Perform correction processing on the hyperspectral image data to obtain corrected image data;

[0009] S3: Dimensionality reduction of the corrected image data is performed using an independent component analysis algorithm to obtain an independent component image;

[0010] S4: Sampling is performed on the independent component images to determine the training samples;

[0011] S5: Train the U-Net model using training samples;

[0012] S6: Input the entire independent component image into the trained U-Net model and output the classification result image;

[0013] S7: Based on the classification result map, assess the operating status of the photovoltaic modules and output a photovoltaic module inspection report.

[0014] A second aspect of this invention provides a drone aerial photography system for inspecting photovoltaic modules, comprising: a processor and a memory;

[0015] The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the drone aerial photography photovoltaic module detection method as described in the first aspect.

[0016] A third aspect of the present invention provides a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the drone aerial photography photovoltaic module detection method described in the first aspect.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0018] In this embodiment of the invention, by using hyperspectral image data and performing dimensionality reduction using an independent component analysis algorithm, interference factors such as ambient temperature, illumination conditions, and surface reflection can be effectively eliminated, significantly reducing the false detection and false negative rates. The rich information provided by hyperspectral data makes it possible to detect defects such as hot spots and microcracks in their early stages or inside the module, improving the depth and sensitivity of detection. By training the U-Net model to automatically classify the entire independent component image, the entire process from data acquisition to condition assessment is automated, greatly reducing the reliance on the experience of professional personnel. While ensuring detection accuracy, it significantly improves the inspection efficiency of photovoltaic modules. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0020] Figure 1 This is a flowchart illustrating a method for detecting photovoltaic modules using drone aerial photography, as provided in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of a drone aerial photography photovoltaic module inspection system provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope 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 should fall within the scope of protection of the present invention.

[0023] The method for detecting photovoltaic modules by drone aerial photography provided by the present invention will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0024] Reference manual attached Figure 1 The diagram shows a flowchart of a method for detecting photovoltaic modules by drone aerial photography according to an embodiment of the present invention.

[0025] This invention provides a method for detecting photovoltaic modules using drone aerial photography, which may include the following steps:

[0026] S1: Acquire hyperspectral image data of photovoltaic modules.

[0027] Specifically, the drone performs overlapping data acquisition along a preset flight path to ensure seamless image coverage and continuous band coverage, providing high-quality data for subsequent analysis. Attitude, GPS location information, and illumination intensity data are recorded during acquisition for geometric and radiometric correction.

[0028] Among them, photovoltaic modules, commonly known as solar panels, are the core power generation units that directly convert sunlight into electrical energy.

[0029] Hyperspectral images refer to images containing hundreds of continuous, narrow spectral bands that can capture fine spectral information of objects far beyond what is visible to the naked eye.

[0030] S2: Perform correction processing on the hyperspectral image data to obtain corrected image data.

[0031] Among them, the correction process refers to the radiometric and geometric correction of the original hyperspectral image to eliminate the effects of sensor error, uneven illumination and topographic distortion, so as to obtain image data that can truly reflect the spectral reflectance characteristics of ground objects.

[0032] Specifically, the acquired raw hyperspectral data is corrected, including emissivity correction, atmospheric correction and geometric correction, and outliers are removed to eliminate interference from the external environment and equipment.

[0033] In this embodiment of the invention, by correcting the original hyperspectral data, sensor errors, uneven illumination, and environmental interference are effectively eliminated, significantly improving the data quality and reliability of subsequent analysis and detection.

[0034] In one possible implementation, S2 specifically includes sub-steps S201 to S203:

[0035] S201: Perform emissivity correction processing on the hyperspectral image data to obtain a reflectivity image:

[0036] .

[0037] in, Indicates wavelength. Indicates at wavelength Reflectance data from the reflectance image. Indicates at wavelength Radiance at that location Indicates at wavelength The dark current response obtained by measuring the black calibration plate. Indicates at wavelength The radiance measured against a white reference plate. This represents the known reflectivity value of the white reference plate, typically 1 or a known high reflectivity value.

[0038] Emissivity correction processing is a key step in separating and eliminating the influence of surface emissivity from the thermal radiation brightness measured from an object, thereby retrieving the object's true physical temperature (i.e., brightness temperature).

[0039] Specifically, radiance is obtained from sensors.

[0040] S202: Perform geometric correction processing on the reflectance image to obtain a finely calibrated image.

[0041] Geometric correction refers to the process of eliminating geometric distortions in remote sensing images caused by factors such as sensor orientation, terrain undulation, and Earth curvature by establishing a mathematical model, thereby matching the coordinates of ground features in the image with their true geographic coordinates.

[0042] In one possible implementation, S202 specifically includes sub-steps S2021 and S2022:

[0043] S2021: The reflectance image is coarsely corrected using a rational function model to obtain a coarsely corrected image.

[0044] Coarse correction refers to the preliminary geometric correction of remote sensing images using auxiliary data such as satellite ephemeris parameters and sensor attitude, in order to systematically eliminate known sensor and platform deformations, but it is a preprocessing stage that has not yet used ground control points for precise registration.

[0045] In one possible implementation, S2021 specifically includes:

[0046] Establish the mathematical relationship between the coordinates of points in the reflectance image and the geographic coordinates of points on the ground:

[0047] .

[0048] Where n represents the normalization process, This represents the row coordinates of the points in the normalized reflectance image. This represents the coordinates of the points in the normalized reflectance image. , as well as These represent the latitude, longitude, and altitude of the normalized ground point, respectively. , , as well as These represent the first and third order polynomials of the normalized ground point geographic coordinates, the second and third order polynomials of the normalized ground point geographic coordinates, the third and third order polynomials of the normalized ground point geographic coordinates, and the fourth and third order polynomials of the normalized ground point geographic coordinates, respectively.

[0049] Based on mathematical relationships, a coarse correction process is performed on the reflectance image to obtain a coarsely corrected image.

[0050] For example, , , and It is the highest order third polynomial, with For example, its general form is:

[0051] .

[0052] in, Let i represent a function with rational coefficients, and i+j+l≤3.

[0053] In this embodiment of the invention, a rational function model based on a third-order polynomial is established to accurately describe the complex spatial mapping relationship between image coordinates and ground geographic coordinates in a unified mathematical form, providing a high-precision initial transformation basis for subsequent geometric fine correction.

[0054] S2022: By using a generative adversarial network, the coarsely calibrated image is finely calibrated to obtain the finely calibrated image.

[0055] .

[0056] in, Indicates a finely calibrated image. This represents the coarsely corrected image. This represents the mapping function for the generator in a generative adversarial network.

[0057] Generative Adversarial Networks (GANs) are deep learning frameworks in which two neural networks, a "generator" and a "discriminator," compete and evolve together. The generator aims to generate data that is indistinguishable from real data, while the discriminator tries to distinguish between real and fake data. The ultimate goal is for the generator to produce extremely realistic results.

[0058] Fine correction refers to further geometric correction of the image based on coarse correction, using precise ground control points to eliminate residual deformation and achieve higher spatial accuracy.

[0059] Specifically, a generative adversarial network (GAN) is used to learn nonlinear mappings to correct the residuals left by RFM.

[0060] In this embodiment of the invention, by combining coarse correction of rational function models with fine correction of generative adversarial networks, the stability of traditional models and the powerful fitting ability of deep learning to complex distortions are fully utilized, thereby achieving sub-pixel level geometric accuracy while ensuring efficiency.

[0061] Optionally, the loss function used to train the generative adversarial network is specifically:

[0062] .

[0063] Where G represents the generator and D represents the discriminator. This indicates a joint optimization process. This represents the optimal generator obtained through training with the objective function. Describes the adversarial loss objective function. Indicates hyperparameters, This represents the objective function for content consistency loss.

[0064] In this context, the objective function is a scalar function used in mathematical optimization models to measure the quality of a solution. The algorithm finds the optimal decision solution by minimizing or maximizing the value of this function.

[0065] Specifically, hyperparameters It is important to balance the loss of adversarial behavior and the loss of content consistency.

[0066] The specific objective function for adversarial loss is:

[0067] .

[0068] in, D represents a real image. This represents the discriminator's judgment output on the real image. This represents the logarithmic probability that the discriminator correctly identifies the target. This indicates that the expression within the parentheses is calculated over all real image samples. This represents the fake image output by the generator. This represents the discriminator's output indicating its judgment of a forged image. This represents the logarithmic probability that the discriminator identifies a fake image. This indicates that the expected value of the expression within the parentheses is calculated for all coarsely corrected image samples.

[0069] The adversarial loss objective function is the core driving force in generative adversarial networks (GANs), used to ensure the "realism" of the generated results. Its goal is to make the images generated by the generator indistinguishable from real images in terms of data distribution to the discriminator, thereby deceiving the discriminator.

[0070] The specific objective function for content consistency loss is as follows:

[0071] .

[0072] in, This indicates that the expected value of the expression within parentheses is calculated for all coarsely corrected image samples and their corresponding ground truth image samples. Let L be the norm.

[0073] The content consistency loss objective function is a supplementary constraint used to maintain the "content consistency" between the generated result and the input source. It ensures that the generation process does not deviate from the basic structure and details of the input content by comparing the differences between the generated image and the original input image at the pixel level or feature level.

[0074] In this embodiment of the invention, by constructing a composite objective function that combines adversarial loss and content consistency loss, the generated image is ensured to have high visual realism while effectively maintaining structural consistency with the input image, thereby achieving accurate correction of geometric distortion.

[0075] S203: Perform outlier removal on the finely calibrated image to obtain the corrected image data:

[0076] .

[0077] in, Indicates at wavelength Reflectance data of the finely calibrated image. This represents the reflectance data of the output corrected image data.

[0078] Outlier removal refers to the process of identifying and removing extreme data points in a dataset that deviate significantly from the main distribution and may distort the analysis results, using statistical or modeling methods.

[0079] In this embodiment of the invention, by sequentially performing emissivity correction, geometric correction, and outlier removal, sensor errors, environmental noise, and data anomalies are systematically eliminated, significantly improving the physical accuracy and geometric precision of hyperspectral data, and providing a reliable data foundation for subsequent fault detection.

[0080] S3: Dimensionality reduction of the corrected image data is performed using an independent component analysis algorithm to obtain an independent component image.

[0081] Independent Component Analysis (ICA) is a blind source separation technique that aims to recover statistically independent, hidden source signals from multiple mixed signals.

[0082] Dimensionality reduction refers to a data processing technique that projects high-dimensional data into a low-dimensional space through a mathematical transformation while preserving as much of the key information of the original data as possible.

[0083] Independent component image (IC image) refers to the two-dimensional projection of each independent component in the spatial dimension obtained after performing independent component analysis on hyperspectral data, where the gray value of each pixel represents the response intensity of the corresponding independent source signal at that location.

[0084] Specifically, to reduce redundancy between hyperspectral bands and improve computational efficiency, independent component analysis (ICA) is used to reduce the dimensionality of hyperspectral image data. The spectral vector of each pixel is treated as a mixed signal, and independent source signals are separated by maximizing the non-Gaussianity criterion. Several independent components with high variance contribution rates are retained to generate corresponding IC images, thereby enhancing the discriminative power of subsequent features.

[0085] In this embodiment of the invention, blind source separation and dimensionality reduction are performed on the corrected hyperspectral data using an independent component analysis algorithm. This effectively removes data redundancy while preserving key spectral features, providing input images with higher signal-to-noise ratio and more significant features for subsequent fault identification.

[0086] In one possible implementation, S3 specifically includes sub-steps S301 to S304:

[0087] S301: Define the spectral vector of each pixel in the corrected image data as a mixed observation signal vector.

[0088] In a pixel digital image, there is the smallest independent sampling unit with uniform color and brightness values.

[0089] Among them, the spectral vector is a multidimensional feature sequence composed of the reflectance or radiance values ​​of a pixel in all spectral bands arranged in order.

[0090] Among them, the hybrid observation signal vector is the comprehensive measurement signal received by a single pixel in hyperspectral remote sensing, which is a mixture of the spectra of multiple ground objects.

[0091] S302: Combining the maximization of non-Gaussianity criterion, an iterative algorithm is used to obtain the projection vector that maximizes the non-Gaussianity of the projection result.

[0092] The iterative algorithm is as follows:

[0093] .

[0094] Where k represents the number of iterations. This represents the projection vector after normalization in the k-th iteration. Indicates the first The projection vector after the iteration normalization. Let x represent the unnormalized projection vector after the k-th iteration update, and let x represent the mixed observation signal vector. Let x represent the covariance matrix of the mixed observation signal vector. Represents the covariance matrix The inverse matrix, This indicates the expected value of the expression within the parentheses.

[0095] Among them, the maximization of non-Gaussianity criterion is a core separation principle adopted by independent component analysis. It advocates approximating the statistical independence of the source by maximizing the non-Gaussianity (such as kurtosis and negative entropy) of the output signal components.

[0096] Among them, iterative algorithms refer to mathematical programs that gradually approach the optimal solution to a problem by repeatedly executing a series of computational steps and using the feedback of each result.

[0097] In this context, the projection vector refers to the weight vector used to define the projection direction in a linear projection transformation. By performing a dot product operation between the projection vector and the original data vector, the scalar value of the data after projection in that direction can be obtained.

[0098] S303: Using the projection vector, linear projection is performed on the mixed observation signal vector to obtain multiple independent source signal vectors, and independent source signal vectors with variance contribution rates greater than the threshold are retained.

[0099] Linear projection refers to the operation of mapping data points in a high-dimensional space to a low-dimensional subspace through linear transformation. Its core is to achieve dimensionality compression by using a linear combination of weight vectors and original features.

[0100] Among them, the variance contribution rate refers to the proportion of the variance carried by a principal component (or factor) to the total variance of the original data, and is used to measure the explanatory power of the component for the variability of the data.

[0101] Among them, the independent source signal vector refers to the original signal vector estimated and recovered from the mixed observation signal by the blind source separation algorithm, in which each component is statistically independent of the others.

[0102] It should be noted that those skilled in the art can set the threshold value according to actual needs, and this invention does not limit it.

[0103] S304: Generate independent component images based on the retained independent source signal vectors.

[0104] Furthermore, the specific steps of the independent component analysis (ICA) dimensionality reduction method are as follows: Separate statistically independent source signals from the hyperspectral data, using them as the most discriminative spectral features. Treat the spectral vector of each pixel as an observed signal x, and find a demixing matrix W such that the source signal s can be obtained through... To restore the image, the algorithm iteratively finds a projection vector w such that the projection result... The non-Gaussianity of is maximized.

[0105] Furthermore, through multiple iterations, all independent components can be separated. The number L of independent components is recorded, and then calculated... L independent IC images are obtained, thereby achieving dimensionality reduction.

[0106] In this embodiment of the invention, statistically independent source signals are separated from mixed spectral signals by maximizing the non-Gaussianity criterion and iterative optimization, and images are generated by screening the main components. This effectively extracts the hidden defect features in photovoltaic modules and significantly improves the signal-to-noise ratio and accuracy of subsequent fault identification.

[0107] S4: Sample from the independent component images to determine the training samples.

[0108] Specifically, representative pixels are manually selected and labeled on the dimensionality-reduced IC image as training samples. The training samples include target categories (normally functioning photovoltaic modules), fault categories (solar panels with dust accumulation, hot spots, or shaded surfaces), and background categories (roofs, concrete floors, vegetation, roads, and water bodies). These samples will be used to train the classification model.

[0109] In this embodiment of the invention, by extracting representative training samples from independent component images, high-quality labeled data covering various typical features is provided for subsequent model training, effectively ensuring the accuracy and generalization ability of the fault identification model.

[0110] S5: Train the U-Net model using training samples.

[0111] Among them, the U-Net model is an encoder-decoder architecture convolutional neural network designed specifically for biomedical image segmentation. Its core feature is to fuse deep semantic features with shallow positional information through skip connections, thereby achieving pixel-level accurate segmentation.

[0112] Specifically, an improved U-Net network structure is adopted, and the dimensionality-reduced multi-channel IC image is input into the network. The encoder part extracts multi-scale spectral-spatial features, and the decoder part restores spatial resolution through upsampling and skip connections. The classification performance is optimized by using the joint loss function of cross-entropy and Dice. The Adam optimizer is used during model training, and an early stopping mechanism is used to prevent overfitting. Finally, a hyperspectral segmentation model that can distinguish normal components, faulty components and background is obtained.

[0113] In this embodiment of the invention, the U-Net model is trained end-to-end using training samples extracted from independent component images, enabling the model to learn the deep spatial-spectral features of photovoltaic module defects, laying the foundation for achieving pixel-level accurate classification.

[0114] S6: Input the entire independent component image into the trained U-Net model and output the classification result image.

[0115] Among them, the classification result map is a thematic image processed by image analysis and pattern recognition, in which each pixel is assigned a category label to intuitively show the spatial distribution of different land features or targets.

[0116] Specifically, the trained U-Net model is applied to the entire hyperspectral image to classify each pixel. The output is a multi-channel mask image, each corresponding to a different class label. After morphological filtering and denoising, a clear classification result image is obtained.

[0117] In this embodiment of the invention, by inputting the complete independent component image into the trained U-Net model, automatic pixel-level classification of the entire image is achieved, and finally a binary segmentation result image that intuitively displays the spatial distribution of the fault area is generated.

[0118] S7: Based on the classification result map, assess the operating status of the photovoltaic modules and output a photovoltaic module inspection report.

[0119] Among them, the inspection report is a structured document formed based on the inspection and data analysis of equipment, systems or environment. Its core purpose is to objectively record the operating status, identify potential faults or hidden dangers, and provide a basis for maintenance decisions.

[0120] In this embodiment of the invention, by analyzing the fault distribution and statistical characteristics in the classification result diagram, an assessment conclusion on the health status of photovoltaic modules is automatically generated, and finally a standardized inspection report with decision support value is output.

[0121] In one possible implementation, S7 specifically includes sub-steps S701 and S702:

[0122] S701: Based on the classification result map, determine the geographical location and quantity information of photovoltaic modules.

[0123] In one possible implementation, S701 specifically includes S7011 and S7012:

[0124] S7011: Perform connected component analysis and boundary extraction on the results to identify the bounding box and geometric contour of the photovoltaic module.

[0125] Connected component analysis refers to the process of finding all interconnected foreground pixel regions in a binary image and assigning a unique identifier to each region.

[0126] Boundary extraction refers to the operation of obtaining the set of edge pixels of a target object through image processing technology, thereby outlining the object's contour.

[0127] The bounding box refers to the smallest outer rectangle that can completely enclose the target object, and is usually represented by the coordinates of its upper left and lower right corners.

[0128] Among them, the geometric contour refers to a closed curve formed by a series of continuous points representing the shape boundary of the target object, which preserves the geometric structure information of the object's shape.

[0129] S7012: Based on the bounding box and geometric contour, combined with GPS positioning information from drone aerial photography, calculate the geographical location and quantity information of each photovoltaic module.

[0130] GPS positioning information refers to spatial location records obtained through the Global Positioning System, which include data such as latitude and longitude coordinates, altitude, and timestamps.

[0131] In this embodiment of the invention, each photovoltaic module is accurately located through connected component analysis and boundary extraction, and the geographic coordinates of the module are accurately calculated by combining GPS positioning information, providing core data support for generating inspection reports with spatial traceability.

[0132] S702: Based on geographical location, quantity information, and classification result map, assess the operating status of photovoltaic modules and output a photovoltaic module inspection report.

[0133] Specifically, the classification results are analyzed for connectivity and boundaries are extracted to automatically identify the bounding boxes and geometric contours of the photovoltaic modules. Combined with GPS positioning information from drone aerial photography, the geographic coordinates, quantity, and arrangement of each photovoltaic module are calculated, providing spatial basis for subsequent inspection and maintenance.

[0134] Furthermore, based on the proportion and distribution characteristics of faulty pixels in the classification results, the operating status of photovoltaic modules is intelligently assessed. The system automatically outputs an inspection report, which includes: a photovoltaic array distribution map and module number, fault type (such as hot spots, dust accumulation, shading) and severity, geographical location and area of ​​each module, proportion of faulty modules, and status trend analysis. The report is presented in a combination of text and graphics, can be exported as a PDF and uploaded to the operation and maintenance management platform, enabling visualized inspection and long-term health monitoring.

[0135] In this embodiment of the invention, by integrating the fault information in the classification result map with the spatial distribution data of photovoltaic modules, a structured inspection report containing geographical location, quantity statistics and status assessment is automatically generated, thereby achieving accurate diagnosis and efficient management of the power plant's operating status.

[0136] Specifically, drones equipped with hyperspectral imagers are used to conduct aerial photography of photovoltaic power plants, collecting spectral data containing dozens to hundreds of continuous narrow bands. The acquired raw hyperspectral images undergo preprocessing, including emissivity correction, atmospheric correction, and geometric correction, to eliminate environmental interference and outliers, obtaining accurate spectral reflectance information. Independent component analysis (ICA) is then used to reduce the dimensionality of the preprocessed hyperspectral data, removing spectral redundancy and retaining the most discriminative spectral features. The dimensionality-reduced data is then input into a pre-trained U-Net semantic segmentation model. This model classifies each pixel in the image into photovoltaic modules, background features, and fault areas based on spectral features, and further distinguishes different states of the photovoltaic modules. Based on the pixel-level classification results, the system automatically locates and outputs the detection results of the photovoltaic modules, generating a detailed report containing fault type, quantity, and geographical location information.

[0137] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0138] In this embodiment of the invention, by using hyperspectral image data and performing dimensionality reduction processing with an independent component analysis algorithm, interference factors such as ambient temperature, illumination conditions, and surface reflection can be effectively eliminated, significantly reducing false detection and false negative rates. The rich information provided by hyperspectral data makes it possible to detect defects such as hot spots and microcracks in their early stages or inside the module, improving the depth and sensitivity of detection. By training the U-Net model to automatically classify the entire independent component image, the entire process from data acquisition to condition assessment is automated, greatly reducing reliance on the experience of professional personnel and significantly improving the inspection efficiency of photovoltaic modules while ensuring detection accuracy.

[0139] Reference manual attached Figure 2 The diagram shows a schematic representation of a drone aerial photography photovoltaic module inspection system provided in an embodiment of the present invention.

[0140] This invention provides a drone aerial photography photovoltaic module inspection system 20, comprising: a processor 201 and a memory 202;

[0141] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described UAV aerial photography photovoltaic module detection method and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0142] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0143] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0144] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as 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, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. 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 via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0145] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0146] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0147] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0148] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0149] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0150] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0151] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0152] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the above-described UAV aerial photography photovoltaic module detection method and achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for inspecting photovoltaic modules using drone aerial photography, characterized in that, include: S1: Acquire hyperspectral image data of photovoltaic modules; S2: Perform correction processing on the hyperspectral image data to obtain corrected image data; S3: The dimensionality of the corrected image data is reduced by the independent component analysis algorithm to obtain an independent component image; S4: Sample the independent component image to determine the training samples; S5: Train the U-Net model using the training samples; S6: Input the entire independent component image into the trained U-Net model and output the classification result image; S7: Based on the classification result diagram, evaluate the operating status of the photovoltaic module and output a photovoltaic module inspection report.

2. The method for detecting photovoltaic modules by drone aerial photography according to claim 1, characterized in that, S2 specifically includes: S201: Perform emissivity correction processing on the hyperspectral image data to obtain a reflectivity image: ； in, Indicates wavelength. Indicates at wavelength Reflectance data from the reflectance image. Indicates at wavelength Radiance at that location Indicates at wavelength The dark current response obtained by measuring the black calibration plate. Indicates at wavelength The radiance measured against a white reference plate. This represents the known reflectance value of the white reference plate, typically 1 or a known high reflectance value; S202: Perform geometric correction processing on the reflectance image to obtain a finely calibrated image; S203: Perform outlier removal processing on the finely calibrated image to obtain the corrected image data: ； in, Indicates at wavelength Reflectance data of the finely calibrated image. This represents the reflectance data of the output corrected image data.

3. The method for detecting photovoltaic modules by drone aerial photography according to claim 2, characterized in that, S202 specifically includes: S2021: The reflectance image is coarsely corrected using a rational function model to obtain a coarsely corrected image; S2022: The coarsely calibrated image is finely calibrated by using a generative adversarial network to obtain the finely calibrated image. ； in, Indicates a finely calibrated image. This represents the coarsely corrected image. This represents the mapping function for the generator in a generative adversarial network.

4. The method for detecting photovoltaic modules by drone aerial photography according to claim 3, characterized in that, S2021 specifically includes: Establish the mathematical relationship between the coordinates of points in the reflectance image and the geographic coordinates of points on the ground: ； Where n represents the normalization process, This represents the row coordinates of the points in the normalized reflectance image. This represents the coordinates of the points in the normalized reflectance image. , as well as These represent the latitude, longitude, and altitude of the normalized ground point, respectively. , , as well as These represent the first and third order polynomials, the second and third order polynomials, the third and fourth order polynomials of the normalized ground point geographic coordinates, respectively. Based on the mathematical relationship, the reflectance image is coarsely corrected to obtain the coarsely corrected image.

5. The method for detecting photovoltaic modules by drone aerial photography according to claim 3, characterized in that, The loss function used to train the generative adversarial network is as follows: ； Where G represents the generator and D represents the discriminator. This indicates a joint optimization process. This represents the optimal generator obtained through training with the objective function. Describes the adversarial loss objective function. Indicates hyperparameters, This represents the objective function for content consistency loss; The specific objective function of the adversarial loss is as follows: ； in, D represents a real image. This represents the discriminator's judgment output on the real image. This represents the logarithmic probability that the discriminator correctly identifies the target. This indicates that the expression within the parentheses is calculated over all real image samples. This represents the fake image output by the generator. This represents the discriminator's output indicating its judgment of a forged image. This represents the logarithmic probability that the discriminator identifies a fake image. This indicates that the expected value of the expression within the parentheses is calculated over all coarsely corrected image samples; The specific objective function for the content consistency loss is as follows: ； in, This indicates that the expected value of the expression within parentheses is calculated for all coarsely corrected image samples and their corresponding ground truth image samples. Let L be the norm.

6. The method for detecting photovoltaic modules by drone aerial photography according to claim 1, characterized in that, S3 specifically includes: S301: Define the spectral vector of each pixel in the corrected image data as a mixed observation signal vector; S302: Combining the maximization of non-Gaussianity criterion, an iterative algorithm is used to obtain the projection vector that maximizes the non-Gaussianity of the projection result; Specifically, the iterative algorithm is as follows: ； Where k represents the number of iterations. This represents the projection vector after normalization in the k-th iteration. Indicates the first The projection vector after the iteration normalization. Let x represent the unnormalized projection vector after the k-th iteration update, and let x represent the mixed observation signal vector. Let x represent the covariance matrix of the mixed observation signal vector. Represents the covariance matrix The inverse matrix, This indicates the expected value of the expression within the parentheses; S303: Using the projection vector, perform linear projection on the mixed observation signal vector to obtain multiple independent source signal vectors, and retain independent source signal vectors whose variance contribution rate is greater than the threshold. S304: Generate the independent component image based on the retained independent source signal vectors.

7. The method for detecting photovoltaic modules by drone aerial photography according to claim 1, characterized in that, Specifically, S7 includes: S701: Based on the classification result map, determine the geographical location and quantity information of the photovoltaic modules; S702: Based on the geographical location, the quantity information, and the classification result map, evaluate the operating status of the photovoltaic module and output a photovoltaic module inspection report.

8. The method for detecting photovoltaic modules by drone aerial photography according to claim 7, characterized in that, Specifically, S701 includes: S7011: Perform connected component analysis and boundary extraction on the results to identify the bounding box and geometric contour of the photovoltaic module; S7012: Based on the bounding box and the geometric contour, and combined with the GPS positioning information obtained from drone aerial photography, calculate the geographical location and quantity information of each photovoltaic module.

9. A drone-based aerial photography system for inspecting photovoltaic modules, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, and when the programs or instructions are executed by the processor, they implement the steps of the drone aerial photography photovoltaic module detection method as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the drone aerial photography photovoltaic module detection method as described in any one of claims 1 to 8.

Images