Photovoltaic module shadowed defect recognition method and system based on multispectral imaging

Abstract

Embodiments of the present application provide a photovoltaic module shadow defect identification method and system based on multispectral imaging. In the method, through the multispectral image data of the surface of the photovoltaic module, the intrinsic features directly related to the material characteristics are extracted by using feature decomposition technology, and a multi-dimensional intrinsic color space capable of accurately reflecting the material characteristics of the module is constructed. By analyzing the relative relationship between the reflectivities of different wave bands, the adaptive clustering algorithm is used to divide the normal area, the shadow area and the defect area in the multi-dimensional intrinsic color space, and based on the material optical characteristics, the two-level identification strategy is used to identify the real defects in the shadow area, so that the accurate identification of the real defects of the photovoltaic module in the complex shadow environment is realized.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic intelligent operation and maintenance technology, and in particular to a method and system for identifying defects in photovoltaic modules under shadow based on multispectral imaging. Background Technology

[0002] During long-term outdoor operation, photovoltaic modules are prone to various defects such as cracks, hot spots, and aging due to environmental stress, which seriously affect power generation efficiency and system operation safety.

[0003] Traditional inspection methods primarily rely on visible light imaging or infrared thermography. However, in practical applications, it has been found that uneven lighting, caused by shadows from clouds, buildings, or vegetation, leads to abnormal grayscale distribution in the image. This often results in shadowed areas being misidentified as defects or masking true defect features, causing numerous false alarms and missed detections. Furthermore, most existing detection methods classify and identify defects based on the apparent texture features or grayscale statistical features of the image, lacking in-depth analysis of the material's inherent optical properties, making it difficult to achieve robust and reliable defect identification under complex and variable outdoor lighting conditions. Summary of the Invention

[0004] This invention provides a method and system for identifying defects in photovoltaic modules under shadow based on multispectral imaging, in order to solve existing problems.

[0005] This invention provides a method for identifying defects in photovoltaic modules under shadow based on multispectral imaging, comprising: Acquire multispectral image data of the photovoltaic module surface; based on the multispectral image data, extract the spectral reflectance curve corresponding to the photovoltaic module pixel by pixel; Principal component analysis was used to perform spectral feature analysis on the spectral reflectance curve to extract the material optical properties of the photovoltaic module; A multi-dimensional intrinsic chromaticity space is constructed based on the optical properties of materials. An adaptive clustering algorithm is used to divide the normal region, shadow region and defect region in the multi-dimensional intrinsic chromaticity space. Based on the optical properties of the material, a two-level identification strategy is used to identify the real defects in the shadow area.

[0006] This invention also provides a photovoltaic module defect identification system based on multispectral imaging under shadow, comprising: The acquisition module is used to acquire multispectral image data of the photovoltaic module surface; and based on the multispectral image data, extract the spectral reflectance curve corresponding to the photovoltaic module pixel by pixel; The analysis module is used to perform spectral feature analysis on the spectral reflectance curve using principal component analysis technology to extract the material optical properties of the photovoltaic module; The clustering module is used to construct a multi-dimensional intrinsic chromaticity space based on the optical properties of materials, and to divide the normal region, shadow region and defect region in the multi-dimensional intrinsic chromaticity space using an adaptive clustering algorithm. An identification module is used to identify real defects in the shadow area based on the optical properties of the material using a two-level identification strategy.

[0007] The present invention also provides an electronic device, which includes at least a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the steps of the method for identifying defects in photovoltaic modules under shadow as described above based on multispectral imaging.

[0008] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for identifying defects in photovoltaic modules under shadow based on multispectral imaging as described above.

[0009] The present invention also provides a computer program product, which includes: computer program code, and when the computer program code is run on a computer, causing the computer to perform the steps of any of the above-described methods for identifying defects in photovoltaic modules under shadow based on multispectral imaging.

[0010] This invention proposes a method for defect identification of photovoltaic modules under shadow conditions based on multispectral imaging. This method utilizes multispectral image data of the photovoltaic module surface and employs feature decomposition technology to extract essential features directly related to material properties, constructing a multi-dimensional intrinsic chromaticity space that accurately reflects the module's material characteristics. By analyzing the relative relationships between reflectivities of different wavelength bands, an adaptive clustering algorithm is used to divide the multi-dimensional intrinsic chromaticity space into normal regions, shadow regions, and defect regions. Based on the material's optical properties, a two-level identification strategy is employed to identify true defects in the shadow region, achieving accurate identification of true defects in photovoltaic modules under complex shadow environments. Attached Figure Description

[0011] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 A flowchart illustrating a method for identifying defects in photovoltaic modules under shadow based on multispectral imaging, provided in an embodiment of the present invention; Figure 2 A flowchart of a photovoltaic module defect identification under shadow based on multispectral imaging provided in this application embodiment; Figure 3 A schematic diagram of a photovoltaic module defect identification device based on multispectral imaging provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0012] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0013] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0014] Example 1: During long-term outdoor operation, photovoltaic modules are prone to various defects such as cracks, hot spots, and aging due to environmental stress, which seriously affect power generation efficiency and system operational safety. Traditional inspection methods mainly rely on visible light imaging or infrared thermography. However, in practical applications, it has been found that when shadows are created by clouds, buildings, or vegetation, uneven illumination can lead to abnormal image grayscale distribution, often misjudging shadowed areas as defects or masking true defect features, resulting in numerous false alarms and missed detections. Existing detection methods are mostly based on the apparent texture features or grayscale statistical features of images for classification and identification, lacking in-depth analysis of the inherent optical properties of materials, making it difficult to achieve robust and reliable defect identification under complex and variable outdoor lighting conditions. With the mature development of multispectral imaging technology, it has become possible to obtain the reflectivity of the module surface at different wavelengths. However, multispectral image data is characterized by high dimensionality and high information redundancy. How to extract essential features directly related to material properties from massive amounts of data and effectively distinguish between illumination changes caused by shadows and the defect features of the material itself remains a significant technical challenge in the field of intelligent photovoltaic inspection.

[0015] Based on this, embodiments of the present invention provide a method for identifying defects in photovoltaic modules under shadow based on multispectral imaging. Figure 1 This is a flowchart illustrating a method for identifying defects in photovoltaic modules under shadow based on multispectral imaging, as provided in an embodiment of the present invention. The process includes the following steps: S101: Acquire multispectral image data of the photovoltaic module surface; based on the multispectral image data, extract the spectral reflectance curve corresponding to the photovoltaic module pixel by pixel; S102: Principal component analysis is used to perform spectral feature analysis on the spectral reflectance curve to extract the material optical properties of the photovoltaic module; S103: Construct a multi-dimensional intrinsic chromaticity space based on the optical properties of materials, and use an adaptive clustering algorithm to divide the normal region, shadow region and defect region in the multi-dimensional intrinsic chromaticity space; S104: Based on the optical properties of the material, a two-level identification strategy is used to identify the real defects in the shadow area.

[0016] The present invention provides a method for identifying defects in photovoltaic modules under shadow based on multispectral imaging. This method solves the problem of misjudgment and missed detection caused by shadow occlusion in complex outdoor environments by traditional visible light or infrared thermal imaging. By deeply exploring the optical properties of the material itself, a robust defect identification mechanism is constructed.

[0017] The method for identifying defects in photovoltaic modules under shadow based on multispectral imaging provided in this invention can be applied to electronic devices, such as PCs or servers.

[0018] In the intelligent operation and maintenance system of modern large-scale photovoltaic power plants, data acquisition is the source of the entire defect identification chain, and its quality directly determines the accuracy of all subsequent analyses. Traditional data acquisition relies on a single image acquisition method, which has extremely limited information dimensions and is easily affected by external environmental interference. These interferences manifest in images as abnormally reduced grayscale values ​​and severely distorted texture and edge features, making them highly susceptible to misjudgment by threshold-based or simple machine learning models as real defects such as cracks, aging, or stains. Therefore, to avoid interference from the external environment and improve the accuracy of data acquisition, this invention constructs a complete imaging quality assurance system. Its core objective is to stably and reliably acquire a set of high-quality multispectral image data that truly reflects the essential properties of photovoltaic module materials under any complex lighting conditions.

[0019] Specifically, multispectral image data of the photovoltaic module surface can be acquired through an environmental perception module that integrates a distributed spectral sensor, an attitude sensor, and a GPS positioning unit. The distributed spectral sensor can be deployed at multiple locations on the device to monitor the solar spectral power distribution incident on the photovoltaic module surface in real time and from all directions. The distributed spectral sensor can not only measure the total irradiance but also acquire changes in spectral composition, such as the attenuation of blue light on cloudy days or the effects of atmospheric Rayleigh scattering. The attitude sensor can record the device's pitch, roll, and yaw angles. The GPS positioning unit is used to determine location information, accurately binding each frame of multispectral image data to a geospatial coordinate system, facilitating subsequent data management and historical comparative analysis.

[0020] In one possible implementation, the electronic device can analyze the acquired multispectral image data to determine the spectral reflectance curve corresponding to the photovoltaic module.

[0021] In another possible implementation, the original multispectral image data may contain some systematic errors or environmental noise pollution. The electronic device can first clean and preprocess the acquired multispectral image data, and then analyze the processed multispectral image data to determine the spectral reflectance curve corresponding to the photovoltaic module.

[0022] In both methods described above, for each pixel in the multispectral image data, the reflectance values ​​of that pixel across all bands are arranged in wavelength order to form a continuous spectral reflectance curve. This spectral reflectance curve represents the response function of the material corresponding to that pixel to electromagnetic waves of different wavelengths under specific lighting conditions, containing rich material information. However, the original spectral reflectance curve is highly susceptible to fluctuations in ambient light intensity; that is, the absolute reflectance measured under strong light and weak light will differ significantly for the same intact component, but this does not change the optical properties of the material itself. Based on this, after determining the spectral reflectance curve corresponding to each pixel, electronic devices can standardize the spectral reflectance curve to eliminate the interference of the external variable of light intensity.

[0023] For example, methods such as maximum value normalization or mean centering can be used to scale each spectral curve to a uniform numerical range or make its mean zero, thereby preserving the relative reflectance characteristics that are only related to the material composition and structure.

[0024] Since the spectral reflectance curves obtained by electronic devices constitute a high-dimensional and redundant dataset, principal component analysis (PCA) is introduced in this invention to extract the most discriminative essential features. The electronic device can use PCA to perform eigenvalue decomposition on the covariance matrix formed by all spectral curves, identifying the principal components that best explain the data variance, including but not limited to the first, second, third, and fourth principal components. These four principal components have been verified to be sufficient to characterize the key optical properties of photovoltaic modules. The first principal component characterizes the overall reflectance properties of the material, reflecting the main variation patterns of the spectral data. The second principal component reflects the absorption characteristics of the material in specific wavelength bands, used to distinguish the compositional differences between different materials, such as absorption valleys or reflection peaks in the near-infrared region. These absorption characteristics are closely related to the aging degree of the ethylene-vinyl acetate copolymer (EVA) film and the type of backing material. The third principal component describes the structural characteristics of the material, reflecting the influence of the material's surface state and internal structure, such as the influence of microcracks, stains, or internal interlayer interface states on light scattering. The fourth principal component contains information on the material's subtle defects, enabling the identification of spectral changes caused by early defects, such as subtle spectral perturbations caused by microcracks or local delamination.

[0025] Electronic devices reduce the complexity of high-dimensional spectral reflectance curves to a set of material optical properties of photovoltaic modules with clear physical meanings and no redundancy by performing layer-by-layer analysis of the above four principal components.

[0026] After successfully extracting the optical properties of the photovoltaic module, the electronic device can organize and interpret these properties within a unified framework to distinguish between three distinct regional states: normal, shadow, and defect.

[0027] Specifically, electronic devices can construct a multidimensional intrinsic chromaticity space based on the optical properties of materials, thereby mapping the extracted spectral features to a unified intrinsic chromaticity space and establishing a complete feature representation of different states of photovoltaic modules. This multidimensional intrinsic chromaticity space is not a traditional Red-Green-Blue (RGB) or Hue-Saturation-Value (HSV) color space, but an abstract mathematical space spanned by the optical properties of materials as coordinate axes. In this multidimensional intrinsic chromaticity space, each pixel is mapped to a high-dimensional vector, and its position is entirely determined by its intrinsic material optical properties, independent of external illumination.

[0028] Regardless of whether a normal component area is covered by shadow, as long as its material is undamaged, its projected position in the multi-dimensional intrinsic chromaticity space should be highly clustered in a specific region. Similarly, real cracks or hot spots will also form distinguishable clusters in this space due to their unique spectral responses. To achieve this semantic segmentation, electronic devices can employ an adaptive clustering algorithm to divide feature points into normal components, shadowed areas, and various defect regions. The clustering parameters are automatically optimized based on the data distribution characteristics to ensure the accuracy of the classification results.

[0029] Specifically, electronic devices can employ an adaptive clustering algorithm to statistically analyze the distribution density, shape, and spacing of all feature points in space, automatically inferring the optimal number of clusters and initial cluster centers. During the clustering process, this adaptive clustering algorithm considers not only spectral features but also integrates spatial neighborhood information and local texture features, forming a multi-feature fusion representation. Furthermore, electronic devices can dynamically adjust the relative importance of spectral, spatial, and texture features in clustering decisions based on currently collected environmental parameters and the specific type of photovoltaic modules. Environmental parameters include, but are not limited to, illumination uniformity and atmospheric transparency, while the specific type of photovoltaic modules includes, but is not limited to, monocrystalline silicon, polycrystalline silicon, passivated emitter and rear cell (PERC). Through this intelligent feature fusion and weight allocation, the adaptive clustering algorithm can generate a robust classification decision boundary, accurately dividing feature points in the multi-dimensional intrinsic chromaticity space into normal regions, shadow regions, and various defect regions.

[0030] For example, in heavy shadows, texture features may be severely distorted, and electronic devices can automatically reduce their weight and rely more on robust spectral features; conversely, in well-lit conditions but with complex stains on the component surface, electronic devices can increase the weight of texture features to aid in judgment.

[0031] Although multi-dimensional intrinsic chromaticity spaces and adaptive clustering algorithms can distinguish between normal, shaded, and defective regions of photovoltaic modules, in practical applications, hidden defects may still exist within shaded regions, or sharp illumination gradients at the edges of shadows may be mistaken for cracks. Therefore, to further improve the accuracy of identification, this embodiment of the invention also utilizes a two-level identification strategy based on the material optical properties of the photovoltaic module to identify true defects within the identified shaded regions.

[0032] Specifically, this invention further proposes a collaborative defect identification and classification method comprising two specific identification stages. In the first stage, shadow region identification, the electronic device calculates the Euclidean distance between the area to be tested and the shadow feature template in the feature space, and combines this with the spectral feature similarity obtained from principal component analysis to accurately locate the shadow region. In the second stage, real defect identification, the electronic device analyzes the spectral response characteristics of the material within the shadow region, and combines this with the spatial distribution patterns of texture features and edge features to effectively distinguish between pseudo-defects caused by shadows and real defects.

[0033] Specifically, in the first level of identification, the electronic device can evaluate the spectral similarity of each pixel within the shadow area. Simultaneously, by combining spatial context information, it detects the presence of false edges or texture abrupt changes caused by shadow boundaries. By setting reasonable thresholds, the electronic device effectively filters out these interfering signals caused by pure illumination variations, thus obtaining the true shadow area. Within this true shadow area, any residual anomalous signals are highly likely to originate from genuine defects.

[0034] In the second level of identification, electronic devices can perform in-depth analysis of real shadow areas. For defects with obvious geometric shapes, such as cracks, the electronic devices extract their spatial and textural features, focusing on detecting the presence of linear structures with high directional consistency that span multiple pixels, and analyzing whether the continuity of this structure is disrupted under shadow coverage. Even in low-contrast shadows, real cracks usually still exhibit a certain degree of physical discontinuity, which can be captured by advanced edge detection and line segment connection algorithms. For temperature-related defects such as hot spots, although multispectral imaging itself does not directly measure temperature, the material in the hot spot region undergoes irreversible chemical or physical changes due to local overheating, such as solder ribbon oxidation or cell burnout. These chemical or physical changes significantly alter its reflection or emission characteristics in specific wavelength bands. By analyzing features sensitive to subtle changes, such as the fourth principal component, chemical or physical methods can indirectly infer the spectral fingerprint of abnormal hot spots, thereby identifying them against a shadow background.

[0035] Based on the above embodiments, in this embodiment of the invention, acquiring multispectral image data of the photovoltaic module surface includes: The acquisition device obtains multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path; Based on the environmental parameters, the sensor response model established through standard whiteboard calibration, and the correction parameters, radiometric correction is performed on the multispectral image sequence to obtain the corrected multispectral image sequence. An improved scale-invariant feature matching algorithm is used, combined with Gaussian pyramid multi-scale feature detection and mismatch elimination algorithm, to perform multi-band image registration on the corrected multispectral image sequence, resulting in a spatially aligned multispectral image sequence. The spatially aligned multispectral image sequence is used as multispectral image data.

[0036] In actual inspection operations of photovoltaic power plants, the raw data captured by the acquisition equipment often contains a large amount of noise and distortion introduced by hardware limitations and environmental interference. Without correction, all subsequent analyses will be based on sand and soil. Therefore, in this embodiment of the invention, the acquisition equipment must acquire multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path. This preset scanning path can be pre-configured to ensure that each photovoltaic module can be imaged at the optimal viewing angle and distance, avoiding perspective distortion caused by an excessively large viewing angle or insufficient resolution due to excessive distance.

[0037] As the acquisition device moves along this path, it not only acquires multispectral image sequences but also environmental parameters. For example, a distributed spectral sensor network within the acquisition device can monitor incident spectral irradiance in real time from all directions, an attitude sensor accurately records the device's three-dimensional orientation, and a GPS positioning unit provides centimeter-level positioning information. The electronic device timestamps and synchronizes these environmental parameters and stores them along with the multispectral image sequences.

[0038] Since the response of each channel of a distributed spectral sensor to light is not ideally linear and drifts with time and temperature, directly using the original multispectral image sequence for analysis is unreliable. Therefore, this invention employs a dual-protection correction strategy, including but not limited to radiometric and geometric corrections, to eliminate systematic errors and deformation effects. A feature matching algorithm is used to achieve accurate registration of multi-band images, providing a high-quality data foundation for subsequent analysis.

[0039] Specifically, before each task begins, the operator places a standard whiteboard with known reflectivity near the area to be tested and images it using the same equipment. By comparing the whiteboard's response value in the image with its standard reflectivity, the electronic equipment can establish an accurate radiometric calibration curve for each band, i.e., a sensor response model. Based on environmental parameters, this sensor response model established through calibration with the standard whiteboard, and correction parameters, the electronic equipment can perform radiometric correction on the multispectral image sequence to obtain the corrected multispectral image sequence.

[0040] Furthermore, in this embodiment of the invention, the electronic device may also have a built-in machine learning model trained based on massive historical calibration data. This machine learning model can fine-tune the calibration curve according to the environmental parameters input in real time to compensate for dynamic factors that cannot be eliminated by a single whiteboard calibration, such as atmospheric path radiation, thereby achieving automatic compensation for sensor characteristics and optical attenuation.

[0041] Since multispectral image sequences are achieved through spectral or time-division methods, there will inevitably be slight spatial misalignments between multispectral images of different bands. While these misalignments may seem insignificant at the pixel level, they can be fatal for tasks requiring pixel-by-pixel analysis of spectral curves. Therefore, this invention employs an improved scale-invariant feature matching algorithm, combined with Gaussian pyramid multi-scale feature detection and mismatch removal algorithms, to perform multi-band image registration on the corrected multispectral image sequence, resulting in a spatially aligned multispectral image sequence.

[0042] Specifically, in the matching stage, the electronic device employs bidirectional nearest neighbor matching to improve robustness and introduces the RANSAC algorithm to remove mismatched points. The electronic device uses bidirectional nearest neighbor matching to fit a global geometric transformation model by repeatedly sampling a small number of matching points, and uses the RANSAC algorithm to decisively remove matching points that deviate too much from the model. Finally, the images of all bands are accurately transformed to the same spatial coordinate system, resulting in a spatially aligned multispectral image sequence. The electronic device uses this spatially aligned multispectral image sequence as the multispectral image data for extracting the spectral reflectance curves corresponding to the photovoltaic modules.

[0043] Based on the above embodiments, in this embodiment of the invention, acquiring a multispectral image sequence and environmental parameters of the photovoltaic module surface using a data acquisition device according to a preset scanning path includes: Based on the image histogram of historical multispectral image sequences, exposure parameters are optimized to obtain optimized exposure parameters. Based on optimized exposure parameters, the acquisition device obtains multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path.

[0044] In complex and variable outdoor lighting conditions, fixed exposure parameters almost inevitably lead to overexposure or underexposure in some images, especially on photovoltaic modules where there are both areas of strong direct sunlight and areas of deep shadow. Overexposure manifests as the loss of highlight details in multispectral image data, while underexposure manifests as the smearing of dark area information in multispectral image data.

[0045] Based on this, in order to ensure that the acquired multispectral image sequence has a sufficient but not overloaded dynamic range in all regions, this embodiment of the invention proposes a dynamic exposure optimization mechanism based on historical image histogram feedback.

[0046] Specifically, before officially starting the scanning of the photovoltaic modules, the electronic device performs a quick preview scan, or uses a sequence of historical multispectral images collected during the same time period and at a similar geographical location from the previous inspection as a reference. The electronic device then performs in-depth analysis of the histograms of these reference images. The histogram is a statistical chart of image brightness distribution, showing the concentrated areas of pixel values. An ideal exposure state should have the main body of the histogram distributed in the mid-tone area, neither too close to the left nor too close to the right. If the main body of the histogram is close to the left, the exposure is underexposed; if the main body is close to the right, the exposure is overexposed. The electronic device also has a built-in intelligent analysis module that calculates several statistical parameters of the current histogram, such as the mean, variance, peak position, and tail length at both ends. If the intelligent analysis module detects widespread highlight clipping, i.e., a large number of pixel values ​​are piled up near the maximum value, the electronic device will automatically calculate an exposure reduction; conversely, if the image is generally dark, the electronic device will increase the exposure. This adjustment is not a simple global increase or decrease, but rather a differentiated processing based on the spectral response characteristics of different bands. For example, the near-infrared band typically receives more energy than the visible light band, so it may require a shorter exposure time. After completing this series of intelligent extrapolations based on historical multispectral image sequences, the electronic device obtains a set of exposure parameters optimized for the current specific lighting scene, including but not limited to shutter speed, aperture size, and ISO sensitivity.

[0047] When the acquisition device officially performs its task according to the preset scanning path, it will acquire multispectral image sequences and environmental parameters of the photovoltaic module surface based on the optimized exposure parameters and the preset scanning path. This ensures that no matter how the environment changes, each frame of multispectral image can retain useful information to the maximum extent within its dynamic range, providing high-quality input for subsequent radiometric correction and feature extraction.

[0048] Specifically, the intelligent analysis module optimizes exposure parameters based on the image histogram of a historical multispectral image sequence. To obtain the optimized exposure parameters, the electronic device can take the image histogram of a single frame of a historical multispectral image from the sequence as input and output the exposure parameters of the next frame. The core of the intelligent analysis module is a predefined optimization objective function, such as maximizing the image's information entropy or minimizing the proportion of saturated pixels at either end of the histogram. The intelligent analysis module calculates the KL divergence between the current histogram and an ideal histogram (such as a uniform or Gaussian distribution) and uses this as an error signal. Then, a simple proportional-integral (PI) controller converts the error signal into an adjustment amount for the exposure value (EV). The electronic device optimizes the exposure parameters based on this adjustment to obtain the optimized exposure parameters.

[0049] Based on the above embodiments, in this embodiment of the invention, the step of using principal component analysis to perform spectral feature analysis on the spectral reflectance curve and extracting the material optical properties of the photovoltaic module includes: The spectral reflectance curve is standardized to eliminate the influence of changes in ambient light and obtain the essential spectral characteristics directly related to the material properties. Principal component analysis (PCA) is used to decompose the essential spectral features into multiple principal component features. The material optical properties of the photovoltaic module are obtained by performing hierarchical analysis on the multiple principal component features.

[0050] The absolute intensity of ambient light is a significant source of interference, causing the same intact photovoltaic module to exhibit drastically different absolute reflectance values ​​at different times and under different weather conditions. If the raw spectral reflectance curve is used directly for analysis, electronic devices will be unable to distinguish whether a low reflectance area is due to shadowing or material degradation.

[0051] Based on this, in this embodiment, after obtaining the spectral reflectance curve corresponding to the photovoltaic module, the electronic device performs standardization processing on the spectral reflectance curve to eliminate the influence of changes in ambient light and obtain the essential spectral characteristics directly related to the material properties. This standardization processing is not a simple normalization, but a mathematical transformation aimed at eliminating the influence of multiplicative illumination factors. Common methods include, but are not preferred, dividing each spectral curve by its maximum value across all bands, or dividing it by its integral area, which represents the total reflected energy.

[0052] After standardizing the spectral reflectance curve, the shape of the spectral reflectance curve is preserved, while its overall amplitude is normalized, so that the curve only reflects the material's relative reflectivity to light of different wavelengths, which is the material's inherent property.

[0053] Subsequently, the electronic device employs principal component analysis (PCA) to decompose the essential spectral features, obtaining multiple principal component features. PCA identifies the directions with the largest variance in the data and projects the data onto these orthogonal directions.

[0054] Specifically, the electronic device performs covariance analysis on the data matrix composed of all essential spectral features, calculating its eigenvectors and eigenvalues. The magnitude of the eigenvalue represents the proportion of data variance explained by the corresponding principal component. The electronic device extracts N principal components, accurately mapping each pixel from the original high-dimensional spectral space to a low-dimensional principal component feature space. The dimension of the high-dimensional spectral space is the number of bands, and the dimension of the principal component feature space is N. Each principal component in this feature space carries a specific physical meaning. The electronic device performs hierarchical analysis on these principal components. This hierarchical analysis, based on a deep understanding of the optical behavior of photovoltaic materials, re-assigns concrete physical meaning to abstract mathematical characteristics, ultimately obtaining a set of material optical properties that can comprehensively and accurately describe the current state of the photovoltaic module.

[0055] Based on the above embodiments, in this embodiment of the invention, the step of performing hierarchical analysis on the multiple principal component features to obtain the material optical properties of the photovoltaic module includes: A first-level analysis is performed on the first principal component of the multiple principal component features to obtain the material reflectance properties that reflect the spectral data variation pattern. A second-level analysis is performed on the second principal component of the multiple principal component characteristics to obtain the band absorption characteristics used to distinguish the differences in material composition; A third-level analysis is performed on the third principal component of the multiple principal component features to obtain material structural features that reflect the influence of the material surface state and internal structure. A fourth-level analysis is performed on the fourth principal component of the multiple principal component features to obtain information on material defects that cause spectral changes; The optical properties of the photovoltaic module are determined based on the material's reflection characteristics, absorption characteristics in the specified band, structural features, and defect information.

[0056] In this embodiment of the invention, the material optical properties of the photovoltaic module are obtained by performing hierarchical analysis on multiple principal component features, including but not limited to performing a first-level analysis on the first principal component of the multiple principal component features, performing a second-level analysis on the second principal component of the multiple principal component features, performing a third-level analysis on the third principal component of the multiple principal component features, and performing a fourth-level analysis on the fourth principal component of the multiple principal component features.

[0057] Specifically, the first principal component (PPC) is the direction with the highest variance contribution, and its value directly reflects the overall reflectivity of the region corresponding to the pixel. For a brand-new monocrystalline silicon module, its PPC score is usually high and evenly distributed; however, as dust accumulates or EVA yellows, the overall reflectivity decreases, and its score also decreases accordingly. Therefore, electronic devices perform first-level analysis on the first principal components of multiple principal component features to obtain the material reflectivity characteristics that reflect the spectral data variation patterns. These material reflectivity characteristics are the primary indicators for evaluating the cleanliness of the module and the aging degree of the encapsulation material.

[0058] The second principal component is used to represent subtle fluctuations in the spectral curve at specific wavelengths. For example, a healthy EVA film has a characteristic absorption valley in the near-infrared band, but the depth and location of this absorption valley change after EVA undergoes aging reactions such as acetic acid precipitation. Similarly, different types of backsheet materials also have their unique spectral "fingerprints." Therefore, electronic devices perform a second-layer analysis of the second principal component characteristics of multiple principal components to obtain the band absorption characteristics used to distinguish differences in material composition. These band absorption characteristics can be used to differentiate material composition, diagnose the chemical state of encapsulation materials, and so on.

[0059] The third principal component is used to represent the light scattering behavior. A smooth, intact solar cell surface produces relatively regular specular reflections, while the presence of microcracks, hidden cracks, or surface contamination leads to a more complex and diffuse light scattering pattern. This change in scattering characteristics is clearly reflected in the value of the third principal component. Therefore, electronic devices perform third-level analysis on the third principal components of multiple principal component features to obtain material structure features that reflect the influence of the material's surface state and internal structure. These material structure features are used to evaluate the physical integrity of the component surface and the bonding state between internal layers.

[0060] The fourth principal component is used to represent early defect signals that are most easily overlooked by conventional detection methods. For example, a newly nascent micron-sized crack may not yet cause a significant degradation in electrical performance or a hot spot, but it has already altered the stress distribution and light propagation path in the local area, thus leaving a trace on the spectrum. Therefore, electronic devices perform fourth-level analysis on the fourth principal component of multiple principal component features to obtain information on material defects that cause spectral changes.

[0061] Electronic devices can integrate the analytical results at four levels to obtain the material optical properties of photovoltaic modules. That is, electronic devices determine the material optical properties of photovoltaic modules based on material reflection characteristics, band absorption characteristics, material structural features, and material defect information.

[0062] For example, electronic devices can construct a four-dimensional panoramic view of the health status of photovoltaic modules based on material reflection characteristics, wavelength absorption characteristics, material structural features, and material defect information. This four-dimensional panoramic view can provide rich information about defect types, locations, severity, and even potential causes.

[0063] In this embodiment of the invention, the layered analysis relies on the fundamental principles of materials optics. The first principal component is related to Fresnel's law of reflection, and its variation is primarily influenced by the material's refractive index and surface roughness. The second principal component is related to molecular vibrations and electronic transitions; specific chemical bonds produce absorption in specific infrared bands. The third principal component is related to Mie scattering theory; when particles or defects with a wavelength comparable to the incident light exist on or within the material's surface, the intensity and angular distribution of the scattered light change. The fourth principal component is related to the material's microscopic stress field and defect state density; these microscopic changes affect light propagation by altering the local dielectric constant.

[0064] Based on the above embodiments, in this embodiment of the invention, the construction of a multi-dimensional intrinsic chromaticity space based on the optical properties of materials, and the division of normal regions, shadow regions, and defect regions in the multi-dimensional intrinsic chromaticity space using an adaptive clustering algorithm, includes: The optical characteristics of the material are mapped onto a unified multidimensional intrinsic chromaticity space to form a distribution of feature points characterizing different states of the photovoltaic module. Based on the distribution pattern of feature points in the multidimensional intrinsic chromaticity space, an optimal classification decision boundary is constructed. Based on the environmental parameters and the type information of the photovoltaic module, and combined with the spectral features, spatial features and texture features in the multispectral image data, the weight coefficients of each feature are optimized through an adaptive weight adjustment mechanism to obtain a weight-optimized multi-feature fusion representation; Based on the optimal classification decision boundary and the multi-feature fusion representation after weight optimization, an adaptive clustering algorithm is used to cluster feature points in the multidimensional intrinsic chromaticity space to obtain normal regions, shadow regions, and defect regions.

[0065] In this embodiment of the invention, the electronic device constructs an intelligent, multi-dimensional intrinsic chromaticity space based on the optical properties of the material, and uses an adaptive clustering algorithm in the multi-dimensional intrinsic chromaticity space to achieve precise division of regions.

[0066] Specifically, the electronic device maps the optical features of the material onto a unified multidimensional intrinsic chromaticity space, forming a distribution of feature points characterizing different states of the photovoltaic module. For example, the electronic device uses the material optical properties corresponding to each pixel as coordinates and projects them into a four-dimensional intrinsic chromaticity space. In this multidimensional intrinsic chromaticity space, normal regions, shadow regions, and defect regions are transformed into point cloud clusters with different distribution patterns. The point cloud clusters in normal regions are typically dense and compact because they share similar material properties; the point cloud clusters in shadow regions, although different in absolute brightness, have positions in the intrinsic space that highly overlap with normal regions because their material remains unchanged, although they may appear slightly diffuse due to a reduced signal-to-noise ratio; while the point cloud clusters of true defects deviate from normal clusters due to their unique spectral response, forming independent sub-clusters.

[0067] To draw clear boundaries in an interwoven point cloud, the electronic device first needs to construct an optimal classification decision boundary. This boundary is not preset but dynamically generated by analyzing the distribution patterns of various types of samples in the current batch of data. Furthermore, to improve the robustness and accuracy of clustering, this embodiment of the invention introduces a multi-feature fusion strategy. In addition to the core spectral features, the electronic device also extracts spatial and texture features for each pixel. Spatial features include, but are not limited to, gradients and curvatures with neighboring pixels, while texture features include, but are not limited to, local binary patterns and statistics of the gray-level co-occurrence matrix. These spectral, spatial, and texture features supplement the deficiencies of spectral information from different perspectives.

[0068] However, the importance of spectral, spatial, and textural features is not static. For example, in dense fog, texture features may be blurred, and their weight should be reduced; while spatial continuity features are crucial when detecting small cracks.

[0069] Based on this, the electronic device is also equipped with an adaptive weight adjustment mechanism. This mechanism reads the environmental parameters recorded during the data acquisition phase and the component type information of the object being inspected in real time. Based on this contextual information, a lightweight weight prediction model dynamically calculates the optimal weight coefficients for spectral features, spatial features, and texture features in the current scene. Based on these optimal weight coefficients, a weight-optimized multi-feature fusion representation is obtained.

[0070] Electronic devices, based on the optimal classification decision boundary and weight-optimized multi-feature fusion representation, employ an adaptive clustering algorithm to cluster feature points in a multi-dimensional intrinsic chroma space, obtaining normal regions, shadow regions, and defect regions. For example, an electronic device can, based on the optimal classification decision boundary and weight-optimized multi-feature fusion representation, use an adaptive clustering algorithm to cluster feature points in a multi-dimensional intrinsic chroma space and generate a semantic segmentation map. This semantic segmentation map clearly indicates whether each pixel in the image belongs to a normal region, a shadow region, or a defect region.

[0071] Based on the above embodiments, in this embodiment of the invention, the step of identifying real defects in the shadow region using a two-level identification strategy based on the optical properties of the material includes: Based on the optical properties of the material, texture abruptness or edge artifact interference caused by shadows is identified and eliminated in the shadow area to obtain the true shadow area; Based on the multispectral image data, the spectral features, spatial features, and texture features within the true shadow region are determined; based on the spectral features, spatial features, and texture features within the true shadow region, the true defects within the true shadow region are identified.

[0072] In this embodiment of the invention, the electronic device can identify real defects in shadowed areas based on the optical properties of materials using a two-stage identification strategy. This two-stage identification strategy includes a first-stage shadowed area identification and a second-stage real defect identification. In the first-stage shadowed area identification, the electronic device calculates the Euclidean distance between the area to be tested and the shadow feature template in the feature space, and combines this with the spectral feature similarity obtained from principal component analysis to accurately locate the shadowed area. In the second-stage real defect identification, the electronic device analyzes the spectral response characteristics of the material within the shadowed area, and combines this with the spatial distribution patterns of texture and edge features to effectively distinguish between pseudo-defects caused by shadows and real defects.

[0073] Specifically, based on the optical properties of materials, electronic devices identify and eliminate texture abruptness or edge artifact interference caused by shadows in the shadow area to obtain the true shadow area; based on multispectral image data, determine the spectral features, spatial features, and texture features within the true shadow area; based on the spectral features, spatial features, and texture features within the true shadow area, identify the true defects within the true shadow area.

[0074] Based on the above embodiments, in this embodiment of the invention, identifying real defects in the real shadow region based on the spectral features, spatial features, and texture features within the real shadow region includes: Based on the spatial and texture features within the real shadow area, the continuity of linear features within the real shadow area is detected; based on the continuity of the linear features, the presence of crack defects within the real shadow area is identified. Based on the spectral characteristics within the true shadow region, abnormal temperature distribution characteristics within the true shadow region are identified; based on the abnormal temperature distribution characteristics, it is determined whether hot spot defects exist within the true shadow region.

[0075] This invention provides specific identification logic for the most common defects in photovoltaic modules: cracks and hot spots. Specifically, for crack identification, the electronic device can utilize the unique spatial and textural characteristics of cracks. A crack is essentially a physical break line; regardless of whether it is covered by shadow, it disrupts the continuity of the material. In the multidimensional intrinsic chromaticity space, the spectral characteristics of a region with cracks will differ from the surrounding normal region due to internal air gaps or contaminant filling; this difference exhibits a strong linear pattern in space.

[0076] Based on this, electronic devices can first apply advanced edge detection operators within a clean, shaded area to extract all potential edge points. Then, using Hough transform or more modern line segment detection algorithms, these discrete edge points are connected into continuous line segments. The electronic device evaluates the length, straightness, and consistency of spectral characteristics along these line segments. A true crack defect typically manifests as a relatively long, straight line segment with spectral characteristics significantly different from the normal areas on either side. Even against a low-contrast background caused by shadows, this structural feature can still be captured as long as the physical presence of the crack remains.

[0077] Furthermore, hot spot defects are overheating phenomena caused by excessive local current density. The root cause is the failure of bypass diodes within the solar cell, poor solder joints, or micro-short circuits within the cell itself. Prolonged overheating can burn out the cell grid lines, carbonize the EVA film, and even melt through the backsheet. These physical or chemical changes permanently alter the material composition and structure of the affected area. Electronic devices pay particular attention to the fourth principal component (PPC) because it is most sensitive to this subtle, localized material degradation. Within clean, shaded areas, the electronic device scans for abnormally high or low PPC values. Simultaneously, it analyzes the reflectance ratios of the area across multiple wavelengths, as the spectral curves of burned areas exhibit characteristic distortions. By combining these spectral anomalies with spatial localization, the electronic device can effectively distinguish hot spots from other types of defects or noise.

[0078] Specifically, identifying real defects within the real shadow region based on its spectral, spatial, and texture features includes: Based on the spatial and texture features within the actual shadow area, the continuity of linear structures is detected; if the linear structure remains continuous and has high directional consistency within the shadow area, it is determined that a crack defect exists. Based on the spectral characteristics of the actual shadow area, analyze its reflectance anomalies or absorption characteristics deviations in specific bands; if a spectral anomaly matching the hot spot aging mode is detected, it is determined that there is a hot spot defect.

[0079] Specifically, electronic devices can detect the continuity of linear structures within a real shadow area based on its spatial and textural features. If the linear structure remains continuous and exhibits high directional consistency within the shadow area, a crack defect is identified. Electronic devices can also analyze reflectivity anomalies or absorption deviations in specific wavelength bands based on the spectral characteristics within the real shadow area. If a spectral anomaly matching a hot spot aging pattern is detected, a hot spot defect is identified.

[0080] In one possible implementation, crack defect detection includes, but is not limited to, the following steps: 1. High-pass filtering is applied to the fourth principal component image to enhance the high-frequency linear structure; 2. Apply nonmaximum suppression and double threshold detection to generate a binary edge map; 3. Use the LSD algorithm to detect line segments with sub-pixel precision; 4. For each candidate line segment, extract a narrow-band profile along its normal direction and calculate the spectral difference on both sides of the profile; 5. If the difference exceeds the threshold and the line segment length is greater than the preset minimum value, it is confirmed as a crack. Hot spot detection uses a combination of template matching and anomaly detection.

[0081] Additionally, the electronic device can be configured with a hot spot spectral template library containing typical spectral curves of hot spots of varying severity. For each candidate region, the electronic device calculates the correlation coefficient between its spectral curve and all templates in the library, and takes the maximum value as the matching score. Simultaneously, the electronic device calculates the standard deviation of the fourth principal component of that region from the mean. Only when both the matching score and the standard deviation from the mean exceed their respective thresholds is the region identified as a hot spot defect.

[0082] Furthermore, electronic devices also utilize knowledge of the circuit topology of components for post-processing verification. For example, hot spot defects typically occur in shielded battery strings; if a detected hot spot defect is located in an unshielded area, it is likely a false alarm.

[0083] For example, the electronic device first establishes a spectral characteristic benchmark for the shaded area, and then identifies the actual defects hidden under the shadow by comparing and analyzing the differences in the spectral responses of the shaded and normal areas. For crack defects, the electronic device identifies them by detecting the continuity of their linear characteristics in the shaded area; for hot spot defects, it determines them by analyzing their abnormal temperature distribution characteristics in the shaded area. This method fully utilizes the distribution characteristics of various regions in the feature space, and realizes a complete workflow from shadow identification to defect determination through a two-stage identification process, significantly improving the accuracy of photovoltaic module defect identification in complex shaded environments.

[0084] Figure 2 This application provides a flowchart for defect identification of photovoltaic modules under shadow based on multispectral imaging, as shown in the embodiments. Figure 2 As shown, the process includes: S201: Start the multispectral imaging system and complete equipment initialization and environmental parameter calibration. Establish the sensor response model through standard whiteboard calibration to ensure measurement accuracy. Control the acquisition equipment to scan along a predetermined path, simultaneously acquiring multispectral image sequences and environmental parameters. Perform radiometric and geometric corrections to eliminate systematic errors and deformation effects. Use a feature matching algorithm to achieve accurate registration of multi-band images, providing a high-quality data foundation for subsequent analysis.

[0085] S202: Based on preprocessed multispectral data, spectral reflectance curves are extracted pixel-by-pixel. Data quality assessment and standardization processes eliminate the influence of ambient lighting variations. Principal component analysis is used to decompose spectral features and extract the intrinsic optical properties of the material. Based on this, a multidimensional intrinsic chromaticity space is constructed, and an adaptive clustering algorithm is used to classify normal components, shadowed areas, and defect regions, establishing a complete feature classification system.

[0086] S203: Based on the constructed feature space, shadowed regions are first accurately identified through feature similarity calculation. Then, a two-level identification strategy is implemented within the shadowed regions: first, interfering features such as texture abrupt changes and edge artifacts caused by shadows are eliminated; then, by analyzing the spectral response characteristics of the material and combining the spatial distribution patterns of texture and edge features, the true defects hidden under the shadows are identified. For different types of defects such as cracks, hot spots, and aging, specialized identification methods are used to achieve accurate classification.

[0087] S204: Generates a complete inspection report including defect distribution, type statistics, and severity. Establishes a defect sample database to store characteristic parameters and environmental information. When the sample accumulation reaches a set scale, automatically initiates the model optimization process, continuously improving system performance through data cleaning and parameter updates. Establishes a performance monitoring mechanism to ensure the long-term stability of identification accuracy.

[0088] Compared with the prior art, the embodiments of the present invention have, but are not limited to, the following advantages: 1. This invention provides a high-precision multispectral imaging and intelligent correction system, which constructs a complete imaging quality assurance system through multi-level technological innovation. Specifically, the system includes three core parts: an environmental perception module, an image processing module, and an intelligent control module. The environmental perception module integrates a distributed spectral sensor, an attitude sensor, and a GPS positioning unit for real-time acquisition of shooting environment parameters and spatial location information. The image processing module performs radiometric correction and image registration processing. Radiometric correction establishes a sensor response model through standard whiteboard calibration and trains correction parameters using machine learning methods to achieve automatic compensation for sensor characteristics and optical attenuation. Image registration employs an improved Scale-invariant Feature Transform (SIFT) feature matching algorithm, using Gaussian pyramid multi-scale feature detection and Random Sample Consensus (RANSAC) algorithm to remove mismatched points, achieving sub-pixel-level accurate registration of multi-band images. The intelligent control module is equipped with an exposure control unit that dynamically optimizes exposure parameters based on image histogram analysis to ensure high-quality image data under different shooting conditions.

[0089] 2. This invention provides a spectral feature hierarchical decomposition and essential feature extraction module: First, the raw spectral data is quality-assessed, and outlier data points in the spectral curves are statistically detected to ensure data quality. Then, standardization processing is used to eliminate the influence of ambient light variations, retaining the essential spectral features directly related to material properties. Next, feature decomposition is performed using principal component analysis (PCA) to extract feature vectors characterizing the material's optical properties by analyzing the covariance matrix of the spectral data. Finally, the extracted features are analyzed hierarchically: the first principal component characterizes the overall reflectance properties of the material, reflecting the main variation patterns of the spectral data; the second principal component reflects the absorption properties of the material in specific wavelength bands, used to distinguish the compositional differences between different materials; the third principal component describes the structural features of the material, reflecting the influence of the material's surface state and internal structure; and the fourth principal component contains information on subtle defects in the material, enabling the identification of spectral changes caused by early defects.

[0090] 3. This invention provides a multi-dimensional intrinsic chromaticity space construction and shadow separation model: By constructing a multi-dimensional feature space, the extracted spectral features are mapped to a unified intrinsic chromaticity space, establishing a complete feature representation of different states of photovoltaic modules. During the feature space construction process, the optimal classification decision boundary is established by analyzing the distribution patterns of various samples. In specific implementation, the system uses an adaptive clustering algorithm to divide feature points into normal components, shadow-affected areas, and various defect regions. The clustering parameters are automatically optimized based on the data distribution characteristics to ensure the accuracy of the classification results. To further improve recognition accuracy, the system adopts a multi-feature fusion strategy, combining spectral features, spatial features, and texture features for comprehensive judgment, and dynamically optimizes the feature weight coefficients according to different environmental conditions and component types through adaptive weight adjustment.

[0091] 4. This invention provides a collaborative defect identification and classification module. Based on the feature foundation provided by the construction of a multidimensional intrinsic chromaticity space and the shadow separation model, this invention further proposes a collaborative defect identification and classification method comprising two specific identification stages. In the first stage, shadow region identification, the system calculates the Euclidean distance between the area to be tested and the shadow feature template in the feature space, and combines this with the spectral feature similarity obtained from principal component analysis to accurately locate the shadow region. In the second stage, real defect identification, the system analyzes the spectral response characteristics of the material within the shadow region, and combines this with the spatial distribution patterns of texture and edge features to effectively distinguish between pseudo-defects caused by shadows and real defects. In specific implementation, the system first establishes a spectral feature benchmark for the shadow region, and then identifies the real defects hidden under the shadow by comparing and analyzing the differences in spectral response between the shadow region and the normal region. For crack defects, the system identifies them by detecting the continuity of their linear features in the shadow region; for hot spot defects, it determines them by analyzing their abnormal temperature distribution characteristics in the shadow region. This method fully utilizes the distribution characteristics of various regions in the feature space, and realizes a complete workflow from shadow identification to defect determination through a two-stage identification process, significantly improving the accuracy of photovoltaic module defect identification in complex shadow environments.

[0092] Example 2: Based on the same concept, Figure 3 A schematic diagram of a photovoltaic module defect identification device based on multispectral imaging provided in an embodiment of the present invention includes: The acquisition module is used to acquire multispectral image data of the photovoltaic module surface; and based on the multispectral image data, extract the spectral reflectance curve corresponding to the photovoltaic module pixel by pixel; The analysis module is used to perform spectral feature analysis on the spectral reflectance curve using principal component analysis technology to extract the material optical properties of the photovoltaic module; The clustering module is used to construct a multi-dimensional intrinsic chromaticity space based on the optical properties of materials, and to divide the normal region, shadow region and defect region in the multi-dimensional intrinsic chromaticity space using an adaptive clustering algorithm. An identification module is used to identify real defects in the shadow area based on the optical properties of the material using a two-level identification strategy.

[0093] In one possible implementation, the acquisition module is specifically used for: The acquisition device obtains multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path; Based on the environmental parameters, the sensor response model established through standard whiteboard calibration, and the correction parameters, radiometric correction is performed on the multispectral image sequence to obtain the corrected multispectral image sequence. An improved scale-invariant feature matching algorithm is used, combined with Gaussian pyramid multi-scale feature detection and mismatch elimination algorithm, to perform multi-band image registration on the corrected multispectral image sequence, resulting in a spatially aligned multispectral image sequence. The spatially aligned multispectral image sequence is used as multispectral image data.

[0094] In one possible implementation, the acquisition module is specifically used for: Based on the image histogram of historical multispectral image sequences, exposure parameters are optimized to obtain optimized exposure parameters. Based on optimized exposure parameters, the acquisition device obtains multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path.

[0095] In one possible implementation, the analysis module is specifically used for: The spectral reflectance curve is standardized to eliminate the influence of changes in ambient light and obtain the essential spectral characteristics directly related to the material properties. Principal component analysis (PCA) is used to decompose the essential spectral features into multiple principal component features. The material optical properties of the photovoltaic module are obtained by performing hierarchical analysis on the multiple principal component features.

[0096] In one possible implementation, the analysis module is specifically used for: A first-level analysis is performed on the first principal component of the multiple principal component features to obtain the material reflectance properties that reflect the spectral data variation pattern. A second-level analysis is performed on the second principal component of the multiple principal component characteristics to obtain the band absorption characteristics used to distinguish the differences in material composition; A third-level analysis is performed on the third principal component of the multiple principal component features to obtain material structural features that reflect the influence of the material surface state and internal structure. A fourth-level analysis is performed on the fourth principal component of the multiple principal component features to obtain information on material defects that cause spectral changes; The optical properties of the photovoltaic module are determined based on the material's reflection characteristics, absorption characteristics in the specified band, structural features, and defect information.

[0097] In one possible implementation, the clustering module is specifically used for: The optical characteristics of the material are mapped onto a unified multidimensional intrinsic chromaticity space to form a distribution of feature points characterizing different states of the photovoltaic module. Based on the distribution pattern of feature points in the multidimensional intrinsic chromaticity space, an optimal classification decision boundary is constructed. Based on the environmental parameters and the type information of the photovoltaic module, and combined with the spectral features, spatial features and texture features in the multispectral image data, the weight coefficients of each feature are optimized through an adaptive weight adjustment mechanism to obtain a weight-optimized multi-feature fusion representation; Based on the optimal classification decision boundary and the multi-feature fusion representation after weight optimization, an adaptive clustering algorithm is used to cluster feature points in the multidimensional intrinsic chromaticity space to obtain normal regions, shadow regions, and defect regions.

[0098] In one possible implementation, the identification module is specifically used for: Based on the optical properties of the material, texture abruptness or edge artifact interference caused by shadows is identified and eliminated in the shadow area to obtain the true shadow area; Based on the multispectral image data, the spectral features, spatial features, and texture features within the true shadow region are determined; based on the spectral features, spatial features, and texture features within the true shadow region, the true defects within the true shadow region are identified.

[0099] In one possible implementation, the identification module is specifically used for: Based on the spatial and texture features within the real shadow area, the continuity of linear features within the real shadow area is detected; based on the continuity of the linear features, the presence of crack defects within the real shadow area is identified. Based on the spectral characteristics within the true shadow region, abnormal temperature distribution characteristics within the true shadow region are identified; based on the abnormal temperature distribution characteristics, it is determined whether hot spot defects exist within the true shadow region.

[0100] Example 3: Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Based on the above embodiments, this embodiment of the present invention also provides an electronic device, including a processor 401, a communication interface 402, a memory 403 and a communication bus 404, wherein the processor 401, the communication interface 402 and the memory 403 communicate with each other through the communication bus 404. The memory 403 stores a computer program that, when executed by the processor 401, causes the processor 401 to perform the steps of any of the above-described methods for identifying defects in the shadows of photovoltaic modules based on multispectral imaging.

[0101] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0102] Communication interface 402 is used for communication between the above-mentioned electronic device and other devices.

[0103] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0104] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0105] Example 4: Based on the above embodiments, this invention also provides a computer-readable storage medium storing a computer program, which is processed by the above-described method for identifying defects in photovoltaic modules under shadow based on multispectral imaging.

[0106] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

[0107] The embodiments described in this invention are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0108] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0110] These computer program instructions may also be stored or loaded in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying defects in photovoltaic modules under shadow based on multispectral imaging, characterized in that, include: Acquire multispectral image data of the photovoltaic module surface; based on the multispectral image data, extract the spectral reflectance curve corresponding to the photovoltaic module pixel by pixel; Principal component analysis was used to perform spectral feature analysis on the spectral reflectance curve to extract the material optical properties of the photovoltaic module; A multi-dimensional intrinsic chromaticity space is constructed based on the optical properties of materials. An adaptive clustering algorithm is used to divide the normal region, shadow region and defect region in the multi-dimensional intrinsic chromaticity space. Based on the optical properties of the material, a two-level identification strategy is used to identify the real defects in the shadow area.

2. The method as described in claim 1, characterized in that, The acquisition of multispectral image data of the photovoltaic module surface includes: The acquisition device obtains multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path; Based on the environmental parameters, the sensor response model established through standard whiteboard calibration, and the correction parameters, radiometric correction is performed on the multispectral image sequence to obtain the corrected multispectral image sequence. An improved scale-invariant feature matching algorithm is used, combined with Gaussian pyramid multi-scale feature detection and mismatch elimination algorithm, to perform multi-band image registration on the corrected multispectral image sequence, resulting in a spatially aligned multispectral image sequence. The spatially aligned multispectral image sequence is used as multispectral image data.

3. The method as described in claim 2, characterized in that, The process of acquiring multispectral image sequences and environmental parameters of the photovoltaic module surface using a data acquisition device according to a preset scanning path includes: Based on the image histogram of historical multispectral image sequences, exposure parameters are optimized to obtain optimized exposure parameters. Based on optimized exposure parameters, the acquisition device obtains multispectral image sequences and environmental parameters of the photovoltaic module surface according to a preset scanning path.

4. The method as described in claim 1, characterized in that, The step of using principal component analysis to perform spectral feature analysis on the spectral reflectance curve to extract the material optical properties of the photovoltaic module includes: The spectral reflectance curve is standardized to eliminate the influence of changes in ambient light and obtain the essential spectral characteristics directly related to the material properties. Principal component analysis was used to decompose the essential spectral features into multiple principal component features. The material optical properties of the photovoltaic module are obtained by performing hierarchical analysis on the multiple principal component features.

5. The method as described in claim 4, characterized in that, The step of performing hierarchical analysis on the multiple principal component features to obtain the material optical properties of the photovoltaic module includes: A first-level analysis is performed on the first principal component of the multiple principal component features to obtain the material reflectance properties that reflect the spectral data variation pattern. A second-level analysis is performed on the second principal component of the multiple principal component characteristics to obtain the band absorption characteristics used to distinguish the differences in material composition; A third-level analysis is performed on the third principal component of the multiple principal component features to obtain material structural features that reflect the influence of the material surface state and internal structure. A fourth-level analysis is performed on the fourth principal component of the multiple principal component features to obtain information on material defects that cause spectral changes; The optical properties of the photovoltaic module are determined based on the material's reflection characteristics, absorption characteristics in the specified band, structural features, and defect information.

6. The method as described in claim 2, characterized in that, The construction of a multi-dimensional intrinsic chromaticity space based on the optical properties of materials, and the use of an adaptive clustering algorithm to divide the space into normal regions, shadow regions, and defect regions, includes: The optical characteristics of the material are mapped onto a unified multidimensional intrinsic chromaticity space to form a distribution of feature points characterizing different states of the photovoltaic module. Based on the distribution pattern of feature points in the multidimensional intrinsic chromaticity space, an optimal classification decision boundary is constructed. Based on the environmental parameters and the type information of the photovoltaic module, and combined with the spectral features, spatial features and texture features in the multispectral image data, the weight coefficients of each feature are optimized through an adaptive weight adjustment mechanism to obtain a weight-optimized multi-feature fusion representation; Based on the optimal classification decision boundary and the multi-feature fusion representation after weight optimization, an adaptive clustering algorithm is used to cluster feature points in the multidimensional intrinsic chromaticity space to obtain normal regions, shadow regions, and defect regions.

7. The method as described in claim 1, characterized in that, The method of identifying real defects in the shadow region based on the optical properties of the material, using a two-level identification strategy, includes: Based on the optical properties of the material, texture abruptness or edge artifact interference caused by shadows is identified and eliminated in the shadow area to obtain the true shadow area; Based on the multispectral image data, the spectral features, spatial features, and texture features within the true shadow region are determined; based on the spectral features, spatial features, and texture features within the true shadow region, the true defects within the true shadow region are identified.

8. The method as described in claim 7, characterized in that, The process of identifying real defects in the real shadow region based on spectral, spatial, and texture features within the real shadow region includes: Based on the spatial and texture features within the real shadow area, the continuity of linear features within the real shadow area is detected; based on the continuity of the linear features, the presence of crack defects within the real shadow area is identified. Based on the spectral characteristics within the true shadow region, abnormal temperature distribution characteristics within the true shadow region are identified; based on the abnormal temperature distribution characteristics, it is determined whether hot spot defects exist within the true shadow region.

9. A photovoltaic module defect identification system under shadow based on multispectral imaging, characterized in that, include: The acquisition module is used to acquire multispectral image data of the photovoltaic module surface; and based on the multispectral image data, extract the spectral reflectance curve corresponding to the photovoltaic module pixel by pixel; The analysis module is used to perform spectral feature analysis on the spectral reflectance curve using principal component analysis technology to extract the material optical properties of the photovoltaic module; The clustering module is used to construct a multi-dimensional intrinsic chromaticity space based on the optical properties of materials, and to divide the normal region, shadow region and defect region in the multi-dimensional intrinsic chromaticity space using an adaptive clustering algorithm. An identification module is used to identify real defects in the shadow area based on the optical properties of the material using a two-level identification strategy.

10. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory being used to store program instructions, and the processor being used to execute the computer program stored in the memory to implement the steps of any of the above-described methods for identifying defects in photovoltaic modules under shadow based on multispectral imaging.

Images