Photovoltaic cell stain identification system and method based on multi-modal data fusion

Abstract

The invention provides a photovoltaic cell stain identification system and method based on multi-modal data fusion, and the system comprises a first data processing module, a first photovoltaic cell voltage and current slope model, a second photovoltaic cell voltage and current slope model, a stain discrimination module, a second data processing module, and a first hot spot identification module. According to the invention, rapid detection of hot spots and accurate positioning of faults are realized through two core dimensions: firstly, the output electrical parameter characteristics of the photovoltaic panel are analyzed, and the operation state and potential problems of the photovoltaic panel are judged in combination with environmental parameters and a specific mathematical model; and 2, performing fast Fourier transform processing on the thermal imaging image of the photovoltaic panel, separating frequency domain features corresponding to the hot spots, and finally accurately positioning the fault position of the photovoltaic panel, thereby realizing fast detection of the hot spots of the photovoltaic panel.

Description

Technical Field

[0001] This invention belongs to the technical field of photovoltaic cell fault handling, and particularly relates to a photovoltaic cell stain identification system and method based on multimodal data fusion. Background Technology

[0002] With the large-scale development of the photovoltaic industry, photovoltaic panels are exposed to the outdoor environment for extended periods. This exposure makes them susceptible to localized heat dissipation problems due to dirt (such as dust, bird droppings, and fallen leaves) and inherent defects in the modules (such as microcracks and aging solder joints), leading to the formation of "hot spots." Hot spots not only cause a sudden rise in localized temperature of the photovoltaic panel (far exceeding normal operating temperatures), resulting in a significant decrease in power generation (localized hot spots can reduce module power generation efficiency by more than 30%), but they can also cause aging of the module's encapsulation layer, glass breakage, and even ignite surrounding cables due to high temperatures, posing serious safety hazards and becoming a core issue restricting the efficiency and safety of photovoltaic power plant operation and maintenance.

[0003] Existing photovoltaic panel hot spot detection technology has significant limitations and cannot meet the operation and maintenance requirements of power plants for "precise location, efficient troubleshooting, and low false alarms." Specific shortcomings are as follows: Limitations of single-mode electrical parameter detection method: Traditional methods only judge the operating status by collecting electrical parameters such as output current, voltage and power generation of photovoltaic panels. However, electrical parameters reflect the "overall performance" of the module and cannot locate the location of local hot spots (for example, the hot spot of a module in a string cannot be distinguished by string-level electrical parameters). It is also difficult to distinguish the cause of hot spots (whether it is dirt obstruction or defect of the module itself), which is prone to "misjudgment (such as classifying normal power fluctuations as hot spots)" or "missed judgment (small area hot spots have a weak impact on the overall electrical parameters and are difficult to identify)".

[0004] Limitations of single-modal image detection methods: Some technologies identify hot spots through visible light or infrared thermal imaging, but there are two major problems: First, thermal imaging images are easily affected by environmental noise (such as sensor noise and changes in outdoor lighting), resulting in blurred grayscale differences between hot spots and the background. Traditional spatial domain image processing (such as threshold segmentation) is difficult to effectively separate hot spots from noise. Second, image detection does not take into account the operating electrical parameters of the photovoltaic panel, and cannot verify whether hot spots actually affect power generation performance (for example, some low-temperature hot spots may only be visual anomalies and do not cause electrical performance degradation), resulting in "ineffective investigation" (investigating false hot spots that have no actual impact), increasing operation and maintenance costs.

[0005] Inefficiency of manual inspection: For large-scale photovoltaic power plants (especially distributed power plants and mountain power plants), manual inspection requires checking each photovoltaic panel one by one, which is not only extremely inefficient (only a few hundred modules can be checked per person per day), but also greatly affected by the experience of personnel. Small areas and hot spots with low temperature differences are easily missed, which cannot meet the power plant's need for "real-time monitoring and rapid response".

[0006] In summary, the industry urgently needs a multimodal fusion detection technology that can determine whether there are "functional hot spots" (hot spots that affect power generation) in the module through electrical parameters, and accurately locate the hot spot position through image processing. At the same time, it can solve the problems of "difficult positioning, high misjudgment, and low efficiency" of single-mode technology, and realize efficient detection of hot spots throughout the entire process "from discovery to location, from judgment to source tracing", providing accurate basis for photovoltaic power plant operation and maintenance. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a multimodal fusion method for identifying hot spots on photovoltaic panels. It achieves rapid hot spot detection and precise fault location through two core dimensions: First, it analyzes the characteristics of the output electrical parameters of the photovoltaic panel and combines environmental parameters with a specific mathematical model to determine the operating status and potential problems of the photovoltaic panel; second, it performs Fast Fourier Transform (FFT) processing on the thermal imaging image of the photovoltaic panel to separate the frequency domain features corresponding to the hot spots, and finally accurately locates the fault position of the photovoltaic panel, thereby achieving rapid detection of hot spots on photovoltaic panels.

[0008] The present invention adopts the following technical solution: A method for identifying stains on photovoltaic cells based on multimodal data fusion, characterized by comprising the following steps: S101 fuses the collected physical parameters of photovoltaic cells to obtain multimodal photovoltaic cell data information; S102 outputs operating voltage for extracting multi-mode photovoltaic cell data. No-load output voltage and no-load output voltage Constructing the voltage and current slope model of the first photovoltaic cell: ; S103 output operating voltage for extracting multi-mode photovoltaic cell data information No-load output voltage and no-load output voltage Constructing the voltage and current slope model for the second photovoltaic cell: ; in: Number of individual photovoltaic panels that are shaded; Blocking the battery's operating current; Determine whether the voltage and current slope model of the second photovoltaic cell is less than that of the voltage and current slope model of the first photovoltaic cell. If the condition is met, proceed to the next step; otherwise, return to S102. S201 divides the acquired infrared images of photovoltaic cells to obtain the spatial domain signal of the photovoltaic cell temperature distribution. ; S202 obtains the frequency domain signal of the photovoltaic cell hot spot by spatial domain processing of the temperature-sensing photovoltaic cell using the Fourier method. ; S203 uses a Gaussian filtering method to analyze the frequency domain signal of the photovoltaic cell hot spot. Optimize the acquisition of high-frequency signals of photovoltaic cell hot spots ; S204 uses the inverse Fourier method to analyze the high-frequency signal of the hot spot in the photovoltaic cell. Processing to obtain the spatial domain of the first photovoltaic cell hot spot ; S205 follows the high-frequency signal of photovoltaic cell hotspots. The spatial domain of the photovoltaic cell hotspot corresponding to the hotspot boundary and core region of the temperature abrupt change Generating the second photovoltaic cell hot spot spatial domain ;Right now:

[0009] S206 employs an optimal adaptive threshold method for the hot spot spatial domain of the second photovoltaic cell. The distribution area of ​​the photovoltaic cell hot spot is obtained by binarizing and separating the hot spot area from the background noise.

[0010] Further, in step S201, the acquired infrared image of the photovoltaic cell is divided to obtain the spatial domain signal of the photovoltaic cell temperature distribution. The process includes: By analyzing the spatial domain signal of photovoltaic cell temperature distribution Noise reduction is performed; that is: ; in: Gaussian kernel balance noise reduction and preservation of hotspot details; By using zero padding, the denoised image Fill to And it is an integer power of 2, that is ; By normalizing the grayscale values ​​of the spatial domain signal of the photovoltaic cell temperature distribution, the grayscale values ​​are mapped to the [0, 255] interval to generate a normalized image. Eliminate brightness differences in different shooting environments: ; in, The maximum grayscale value of the image; This is the minimum grayscale value of the image.

[0011] Furthermore, in step S203, the frequency domain signal of the photovoltaic cell hot spot is processed using a Gaussian filtering method. Optimize the acquisition of high-frequency signals of photovoltaic cell hot spots The process includes: Perform 2D-FFT on the filled image to convert the spatial domain thermal imaging into the frequency domain to obtain the high-frequency signal corresponding to the initial hot spot: ; The high-frequency signal corresponding to the initial hot spot is obtained by processing the high-frequency signal corresponding to the initial hot spot through Gaussian high-pass filtering; ; in: After frequency domain centering, the coordinates ) to the frequency domain center Euclidean distance: ; Cutoff frequency; ≥ High-frequency components are retained. < Low-frequency components are suppressed.

[0012] Furthermore, step S204 uses the inverse Fourier method to analyze the high-frequency signal of the photovoltaic cell hotspot. Processing to obtain the spatial domain of the first photovoltaic cell hot spot The process includes: The centered frequency domain signal With high-pass filter Multiplication filters out low-frequency background near the center; ; Restore the original frequency domain position of the filtered frequency domain signal: ; Perform IFFT on the original frequency domain location after filtering, transform it back to the spatial domain, and obtain a complex image containing the target features; .

[0013] Furthermore, step S206 employs an optimal adaptive threshold method for the spatial domain of the hot spot in the second photovoltaic cell. The hot spot distribution region of the photovoltaic cell is obtained by binarizing and separating the hot spot region from the background noise, where: ; in: This is the hot spot region; For the background.

[0014] The present invention can also adopt the following technical solutions: A photovoltaic cell stain identification system based on multimodal data fusion includes a first data processing module, a first photovoltaic cell voltage and current slope model, a second photovoltaic cell voltage and current slope model, a stain discrimination module, a second data processing module, a first hot spot identification module, a second hot spot identification module, a third hot spot identification module, a fourth hot spot identification module, and a hot spot localization module, wherein: The first data processing module is used to fuse the collected physical parameters of photovoltaic cells to obtain multimodal photovoltaic cell data information; The stain discrimination module determines whether the voltage and current slope model of the second photovoltaic cell is less than that of the voltage and current slope model of the first photovoltaic cell. The second data processing module divides the acquired infrared images of photovoltaic cells to obtain the spatial domain signal of the photovoltaic cell temperature distribution; The first hot spot identification module is used to process the temperature-sensing photovoltaic cell spatial domain to obtain the frequency domain signal of the photovoltaic cell hot spot; The second hot spot identification module obtains the high-frequency signal of the photovoltaic cell hot spot by optimizing the frequency domain signal of the photovoltaic cell hot spot using a Gaussian filtering method; The third hot spot identification module obtains the spatial domain of the first photovoltaic cell hot spot by processing the high-frequency signal of the photovoltaic cell hot spot using the inverse Fourier method. The fourth hot spot identification module generates a second photovoltaic cell hot spot spatial domain according to the hot spot boundary and core region corresponding to the photovoltaic cell hot spot spatial domain based on the temperature change of the high-frequency signal of the photovoltaic cell hot spot. The hot spot localization module uses the optimal adaptive threshold method to perform binarization separation of the hot spot region and background noise in the second photovoltaic cell hot spot spatial domain to obtain the photovoltaic cell hot spot distribution region.

[0015] Beneficial effects This invention proposes a multimodal fusion method for identifying hot spots on photovoltaic panels, achieving rapid hot spot detection and accurate fault location through two core dimensions: First, multimodal fusion overcomes the limitations of single-modal technology. This invention utilizes a dual-dimensional approach combining electrical parameters and thermal imaging. By integrating the output electrical parameters of the photovoltaic panel (voltage, current, and slope models) with frequency domain processing of thermal images, it not only identifies "functional hot spots" (real faults affecting power generation efficiency) through electrical parameters but also precisely locates them through images, solving the core problems of traditional single-modal technology: "difficult positioning, high false positives, and many missed detections." Through dual verification of causes and effects, electrical parameter analysis correlates with causes such as dirt shading / module defects, while thermal imaging verifies the degree of temperature anomalies, avoiding "ineffective detection of false hot spots" and "missed detection of small-area hot spots," providing a comprehensive decision-making basis for operation and maintenance.

[0016] Secondly, the accuracy of hot spot recognition is significantly improved. This invention uses quantitative analysis of electrical parameter models to construct a dual voltage and current slope model (k and k'). By comparing the slope differences between shaded and unshaded batteries, the number and area of ​​shaded cells are quantitatively determined, achieving accurate identification of hot spot-related electrical performance anomalies. Image recognition is optimized through frequency domain processing, using 2D-FFT transformation + Gaussian high-pass filter (GHPF) to separate high-frequency features of hot spots from low-frequency noise in the background, solving the problem of weak anti-interference capability of traditional spatial domain processing. Combined with Otsu adaptive threshold segmentation, optimal separation of hot spots and background is achieved by maximizing inter-class variance, improving the recognition rate of small-area, low-temperature difference hot spots. Image preprocessing enhances robustness by eliminating sensor noise, shooting environment brightness differences, and FFT aliasing interference through Gaussian noise reduction, zero filling, and grayscale normalization, ensuring recognition stability under different operating conditions.

[0017] Thirdly, it significantly improves operation and maintenance efficiency and safety. This invention enables rapid detection and precise location, preliminary screening of electrical parameters to narrow down the fault range, and thermal imaging frequency domain processing to quickly output the boundaries, core areas, and coordinates of hot spots. Combined with drone inspections, it can meet the "real-time monitoring and rapid response" requirements of large power plants, replacing inefficient manual inspections; reducing operation and maintenance costs and safety risks, reducing ineffective troubleshooting caused by misjudgments / missed judgments, avoiding safety hazards such as component aging, glass breakage, and fires caused by hot spots, while also reducing power generation loss. Attached Figure Description

[0018] Figure 1 This is a flowchart of a multimodal fusion photovoltaic panel hot spot identification method according to the present invention. Detailed Implementation

[0019] The following is in conjunction with the appendix Figure 1 The present invention will be described in detail as follows: This invention discloses a multimodal fusion method for identifying hot spots in photovoltaic cells. This method integrates photovoltaic cell output electrical parameter analysis with thermal imaging image frequency domain processing technology to achieve rapid hot spot identification and precise fault location. Electrical parameter acquisition and analysis include: collecting environmental parameters of the photovoltaic power station (ambient temperature, wind speed, wind direction, and light intensity) and electrical parameters of equipment such as the BMS and inverter; calculating power generation, output current, and output voltage using theoretical power generation formulas, photovoltaic cell current output models, current characteristic equations, and output voltage formulas; and determining whether the photovoltaic panel has stains or inherent defects based on electrical parameter relationships (such as comparing the slopes of the operating curves of shaded and unshaded cells). Through electrical parameter calculation and model analysis, it is determined whether the photovoltaic panel is in normal operating condition and potential problems such as stains and inherent defects are preliminarily identified. Parameter acquisition: obtaining environmental parameters (light intensity) and equipment electrical parameters (relevant data from the BMS, inverter, remote control system, and relay protection devices).

[0020] This invention calculates theoretical power generation based on environmental parameters:

[0021] in, Theoretical power generation of photovoltaic panels Total irradiance of photovoltaic panels Irradiance under standard conditions Overall system efficiency Photovoltaic panel installation capacity; Photovoltaic cell current output model: Based on parameters such as photocurrent and diode reverse saturation current, the relationship between current, voltage, and temperature is described by nonlinear equations; Photovoltaic cell module current characteristic equation: divided by voltage range ( , Establish a piecewise function to reflect the current variation pattern; Output voltage formula: The total output voltage of the photovoltaic panel is calculated using the voltage and slope parameters of the shaded and unshaded cells.

[0022] I. Power Parameter Acquisition and Analysis The presence or absence of photovoltaic (PV) panels can be determined by collecting their electrical parameters. The testing method is as follows: (a) Photovoltaic panel power generation Based on environmental parameters (illuminance) collected by the photovoltaic power station's environmental monitoring instrument, and by collecting electrical parameters from the BMS and inverter (remote control system, relay protection device), the electrical parameters of the photovoltaic modules are calculated, including power generation. Output current Output voltage .

[0023] 1. Theoretical power generation of a single photovoltaic panel

[0024] judge < If the photovoltaic panel generates less electricity than the design standard, then the power generation will be lower than the design requirement.

[0025] The power generation of photovoltaic panels is calculated based on the following formula:

[0026] Theoretical power generation of photovoltaic panels ; Total irradiance of photovoltaic panels ; Irradiance under standard conditions ; Photovoltaic panel installation power, ; Overall system efficiency.

[0027] 2. Calculation of photovoltaic cell voltage, current, and slope parameters: (1) Formula for the slope of unblocked voltage and current: ; in, Slope of the operating curve of an unobstructed battery; Actual output voltage of a single photovoltaic cell ; The no-load output voltage of a single photovoltaic cell. ; The no-load output voltage of a single photovoltaic cell. ; (2) Formula for the slope of blocking voltage and current: ; in, Slope of the operating curve of the obscured battery; The output operating voltage of the unshaded photovoltaic panel. ; Number of individual photovoltaic panels that are shaded; Unshaded photovoltaic panels operate at high current. Blocking the battery's operating current; The no-load output voltage of a single photovoltaic cell. ; By judgment and The size relationship can be used to determine the number of individual cells that are being blocked. , The smaller the slope, The larger the value, the larger the area of ​​the shaded battery. This helps identify photovoltaic panels with output parameter problems. By judging the electrical parameters of the photovoltaic panel, it is possible to determine whether there are stains or defects in the photovoltaic panel itself.

[0028] Judgment Logic: By comparing the slopes of the operating curves of shaded and unshaded cells, the number and area of ​​shaded cells are determined, thereby identifying abnormal electrical parameters of the photovoltaic panel (related to stains or inherent defects); specifically: S101 fuses the collected physical parameters of photovoltaic cells to obtain multimodal photovoltaic cell data information; S102 outputs operating voltage for extracting multi-mode photovoltaic cell data. No-load output voltage and no-load output voltage Constructing the voltage and current slope model of the first photovoltaic cell: ; S103 output operating voltage for extracting multi-mode photovoltaic cell data information No-load output voltage and no-load output voltage Constructing the voltage and current slope model for the second photovoltaic cell:

[0029] in: Number of individual photovoltaic panels that are shaded; Blocking the battery's operating current Determine whether the voltage and current slope model of the second photovoltaic cell is less than that of the voltage and current slope model of the first photovoltaic cell. If the condition is met, proceed to the next step; otherwise, return to S102. The thermal imaging image processing workflow includes: image preprocessing (Gaussian noise reduction, zero-filling, grayscale normalization), 2D-FFT frequency domain transformation, Gaussian high-pass filtering (GHPF), inverse FFT spatial domain restoration, Otsu adaptive thresholding segmentation, and binarization, achieving effective separation of hot spots from the background and noise. The cutoff frequency of the Gaussian high-pass filter is adaptively adjusted according to the minimum pixel scale of the hot spot, and the threshold is determined by maximizing the inter-class variance, achieving accurate extraction of the hot spot region.

[0030] This invention, based on primary electrical parameter localization, achieves precise hotspot localization through frequency domain processing. It includes: 1. Perform image preprocessing The first step is to smooth sensor noise using a Gaussian filter to balance noise reduction with preservation of hot spot details; The second step is to adjust the image to a size that is a power of 2 by zero padding, so as to avoid aliasing during FFT cyclic convolution. The third step is to normalize the grayscale values ​​and map them to the [0,255] range to eliminate differences in the brightness of the shooting environment.

[0031] The fourth step involves performing 2D-FFT on the preprocessed image through frequency domain conversion and analysis, converting the spatial domain thermal imaging into a frequency domain signal. The low-frequency components correspond to a uniform background, while the high-frequency components correspond to hot spots (regions with abrupt temperature changes).

[0032] The fifth step involves frequency domain filtering using a Gaussian high-pass filter (GHPF) to retain the high-frequency components corresponding to the hot spot while suppressing the low-frequency components of the background and the high-frequency components of noise. The cutoff frequency is adaptively adjusted according to the minimum size of the hot spot.

[0033] 2. Spatial Domain Reconstruction and Feature Enhancement: The first step is to convert the filtered frequency domain signal back to the spatial domain using inverse FFT, and take the complex image modulus as the effective signal. The second step is to find the optimal threshold using the Otsu algorithm to achieve binarization separation of the hot spot area and the background noise.

[0034] Second, the third step involves extracting the bounding box, center coordinates, and average grayscale value (corresponding to temperature value) of the hot spot region to accurately locate the fault point. Specifically: S201 divides the acquired infrared images of photovoltaic cells to obtain the spatial domain signal of the photovoltaic cell temperature distribution. ; S202 obtains the frequency domain signal of the photovoltaic cell hot spot by performing spatial domain Fourier transform on the temperature-sensing photovoltaic cell. ; S203 uses a Gaussian filtering method to analyze the frequency domain signal of the photovoltaic cell hot spot. Optimize the acquisition of high-frequency signals of photovoltaic cell hot spots ; S204 uses the inverse Fourier method to analyze the high-frequency signal of the hot spot in the photovoltaic cell. Processing to obtain the spatial domain of the first photovoltaic cell hot spot ; S205 follows the high-frequency signal of photovoltaic cell hotspots. The spatial domain of the photovoltaic cell hotspot corresponding to the hotspot boundary and core region of the temperature abrupt change Generating the second photovoltaic cell hot spot spatial domain ; S206 employs an optimal adaptive threshold method for the hot spot spatial domain of the second photovoltaic cell. The distribution area of ​​the photovoltaic cell hot spot is obtained by binarizing and separating the hot spot area from the background noise.

[0035] In practice, based on the initial positioning of the photovoltaic panel, thermal imaging images are acquired by a drone, and the Fast Fourier Transform (FFT) is used to achieve accurate identification and positioning of hot spots. The core logic is based on the characteristic of "spatial domain abrupt change - frequency domain high-frequency concentration": the hot spot in the thermal image corresponds to a temperature abrupt change (rapid change in local grayscale in the spatial domain), which is represented as high-frequency components in the frequency domain after FFT transformation. The high-frequency target and low-frequency background are separated by frequency domain filtering, and then the spatial domain features are restored by inverse transform, ultimately achieving accurate positioning of the hot spot.

[0036] Based on the table above, this invention improves the frequency domain separation effect through image preprocessing. Thermal imaging images are susceptible to noise and need to be downgraded first. Optimize image quality: (1) Preprocessing (noise reduction, size alignment) Thermal imaging is subject to sensor noise, which needs to be reduced first to avoid high-frequency noise interfering with hot spot identification. At the same time, the size is aligned with FFT optimization and adjusted to an integer power of 2.

[0037] Gaussian noise reduction: Noise is smoothed using a Gaussian filter; the spatial domain convolution form is...

[0038] in Gaussian kernel ( Balancing noise reduction with preservation of hotspot details; Zero padding: This refers to the noise-reduced image. Fill to And it is an integer power of 2, denoted as To avoid aliasing during FFT circular convolution, fill regions with zeros.

[0039] Gray-level normalization: Maps the gray-level values ​​of an image to the range [0, 255] to generate a normalized image. Eliminate brightness differences in different shooting environments:

[0040] in, The maximum grayscale value of the image; The minimum gray value of the image 2. Spectral Feature Analysis: Low-frequency region (center): corresponds to the uniform gray level of the background, with high amplitude and concentrated energy; High-frequency region (edge): corresponds to abrupt changes in grayscale of the target (such as local overheating caused by hot spots), with low amplitude but significant features; The selection of high-frequency components for thermal imaging targets is concentrated on the spectral radius. The filtering radius for a given region can be determined using the histogram thresholding method.

[0041] 2D-FFT frequency conversion Perform 2D-FFT on the filled image to convert the spatial domain thermal imaging into the frequency domain, and analyze the high-frequency components corresponding to the hot spots:

[0042] 3. Frequency domain filtering: Separating the target from the background Design a high-pass filter to retain high-frequency target characteristics and suppress low-frequency background.

[0043] High-pass filter It retains the high-frequency components corresponding to the hot spot, while suppressing the low-frequency components and noise high-frequency components in the normal area.

[0044] (a) Filter design: Gaussian high-pass filter (GHPF) The frequency domain response of a Gaussian high-pass filter is

[0045] After frequency domain centering, the coordinates ) to the frequency domain center Euclidean distance:

[0046] Cutoff frequency (adaptively adjusted based on hotspot parameters); ≥ High-frequency components are retained. < Low-frequency components are suppressed; Adaptive adjustment: Let the minimum scale of the hotspot be s (pixels), then the spatial frequency corresponding to the hotspot is... (Frequency is inversely proportional to size), cutoff frequency (Mapping spatial frequency to frequency domain distance) (b) Filtering operation The centered frequency domain signal With high-pass filter Multiplying these components yields the filtered frequency domain signal, retaining only the high-frequency components far from the center (grayscale abrupt changes corresponding to hot spots) while filtering out the low-frequency background near the center.

[0047]

[0048] (c) Reverse centralization The filtered frequency domain signal is restored to its original frequency domain position (to counteract the centering in step 2):

[0049] 4. Inverse Fourier Transform and Feature Enhancement Inverse FFT: Performing an inverse FFT (IFFT) on the filtered frequency domain matrix transforms it back into the spatial domain, resulting in a complex image containing the target features.

[0050] For the filtered frequency domain signal Perform 2D-IFFT to recover the spatial domain image (containing only the high-frequency abrupt change regions corresponding to hot spots): ; Since it is a complex number, its magnitude (amplitude) is taken as the effective signal.

[0051] The areas with higher gray values ​​correspond to the boundaries and core regions of hot spots where temperature changes occur abruptly in the original thermal image.

[0052] Contrast enhancement: The Otsu algorithm with adaptive thresholding separates the target feature region from the background noise. right Perform threshold segmentation to separate the hot spot region from the background noise: Adaptive threshold calculation: Using the Otsu thresholding method, the optimal threshold T is found, satisfying:

[0053] in: The inter-class variance after threshold T segmentation is the optimal value for separating hot spots from the background when the inter-class variance is maximized.

[0054] Binarization:

[0055] This is the hot spot region; Background 5. Post-processing: Coordinate output Extract the bounding box, center coordinates, and grayscale mean (corresponding temperature value) of the target area to provide data support for subsequent accurate fault analysis.

[0056] satisfy The coordinate region is the hot spot distribution area, thereby achieving target area positioning.

[0057] Although the present invention has been described above, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many modifications under the guidance of the present invention without departing from the spirit of the present invention, and these modifications are all within the protection scope of the present invention.

Claims

1. A method for identifying stains in photovoltaic cells based on multimodal data fusion, characterized in that, Includes the following steps: S101 fuses the collected physical parameters of photovoltaic cells to obtain multimodal photovoltaic cell data information; S102 outputs operating voltage for extracting multi-mode photovoltaic cell data. No-load output voltage and no-load output voltage Constructing the voltage and current slope model of the first photovoltaic cell: ； S103 output operating voltage for extracting multi-mode photovoltaic cell data information No-load output voltage and no-load output voltage Constructing the voltage and current slope model for the second photovoltaic cell: ； in: Number of individual photovoltaic panels that are shaded; Blocking the battery's operating current; Determine whether the voltage and current slope model of the second photovoltaic cell is less than that of the voltage and current slope model of the first photovoltaic cell. If the condition is met, proceed to the next step; otherwise, return to step S102. S201 divides the acquired infrared images of photovoltaic cells to obtain the spatial domain signal of the photovoltaic cell temperature distribution. ; S202 obtains the frequency domain signal of the photovoltaic cell hot spot by spatial domain processing of the temperature-sensing photovoltaic cell using the Fourier method. ; S203 uses a Gaussian filtering method to analyze the frequency domain signal of the photovoltaic cell hot spot. Optimize the acquisition of high-frequency signals of photovoltaic cell hot spots ; S204 uses the inverse Fourier method to analyze the high-frequency signal of the hot spot in the photovoltaic cell. Processing to obtain the spatial domain of the first photovoltaic cell hot spot ; S205 follows the high-frequency signal of photovoltaic cell hotspots. The spatial domain of the photovoltaic cell hotspot corresponding to the hotspot boundary and core region of the temperature abrupt change Generating the second photovoltaic cell hot spot spatial domain ;Right now: S206 employs an optimal adaptive threshold method for the hot spot spatial domain of the second photovoltaic cell. The distribution area of ​​the photovoltaic cell hot spot is obtained by binarizing and separating the hot spot area from the background noise.

2. The photovoltaic cell stain identification method based on multimodal data fusion according to claim 1, characterized in that, Step S201 involves dividing the acquired infrared image of the photovoltaic cell to obtain the spatial domain signal of the photovoltaic cell temperature distribution. The process includes: By analyzing the spatial domain signal of photovoltaic cell temperature distribution Noise reduction is performed; that is: ； in: Gaussian kernel balance noise reduction while preserving hotspot details; By using zero padding, the denoised image Fill to And it is an integer power of 2, that is ; By normalizing the grayscale values ​​of the spatial domain signal of the photovoltaic cell temperature distribution, the grayscale values ​​are mapped to the [0, 255] interval to generate a normalized image. Eliminate brightness differences in different shooting environments: ； in, The maximum grayscale value of the image; This is the minimum grayscale value of the image.

3. The photovoltaic cell stain identification method based on multimodal data fusion according to claim 2, characterized in that, Step S203 involves using a Gaussian filter to process the frequency domain signal of the photovoltaic cell hotspot. Optimize the acquisition of high-frequency signals of photovoltaic cell hot spots The process includes: Perform 2D-FFT on the filled image to convert the spatial domain thermal imaging into the frequency domain to obtain the high-frequency signal corresponding to the initial hot spot: ； The high-frequency signal corresponding to the initial hot spot is obtained by processing the high-frequency signal corresponding to the initial hot spot through Gaussian high-pass filtering; ； in: After frequency domain centering, the coordinates ) to the frequency domain center Euclidean distance: ； Cutoff frequency; ≥ High-frequency components are retained. < Low-frequency components are suppressed.

4. The photovoltaic cell stain identification method based on multimodal data fusion according to claim 1, characterized in that, Step S204 uses the inverse Fourier method to analyze the high-frequency signal of the photovoltaic cell hot spot. Processing to obtain the spatial domain of the first photovoltaic cell hot spot The process includes: The centered frequency domain signal With high-pass filter Multiplication filters out low-frequency background near the center; ； Restore the original frequency domain position of the filtered frequency domain signal: ； Perform IFFT on the original frequency domain location after filtering to transform it back to the spatial domain, and obtain a complex image containing the target features; 。 5. The photovoltaic cell stain identification method based on multimodal data fusion according to claim 1, characterized in that, Step S206 employs an optimal adaptive threshold method to optimize the spatial domain of the second photovoltaic cell hotspot. The hot spot distribution region of the photovoltaic cell is obtained by binarizing and separating the hot spot region from the background noise, where: ； in: This is the hot spot region; For the background.

6. A photovoltaic cell stain identification system based on multimodal data fusion, characterized in that, The identification system is implemented based on any one of claims 1-5, and includes a first data processing module, a first photovoltaic cell voltage and current slope model, a second photovoltaic cell voltage and current slope model, a stain discrimination module, a second data processing module, a first hot spot identification module, a second hot spot identification module, a third hot spot identification module, a fourth hot spot identification module, and a hot spot positioning module, wherein: The first data processing module is used to fuse the collected physical parameters of photovoltaic cells to obtain multimodal photovoltaic cell data information; The stain discrimination module determines whether the voltage and current slope model of the second photovoltaic cell is less than that of the voltage and current slope model of the first photovoltaic cell. The second data processing module divides the acquired infrared images of photovoltaic cells to obtain the spatial domain signal of the photovoltaic cell temperature distribution; The first hot spot identification module is used to process the temperature-sensing photovoltaic cell spatial domain to obtain the frequency domain signal of the photovoltaic cell hot spot; The second hot spot identification module obtains the high-frequency signal of the photovoltaic cell hot spot by optimizing the frequency domain signal of the photovoltaic cell hot spot using a Gaussian filtering method; The third hot spot identification module obtains the spatial domain of the first photovoltaic cell hot spot by processing the high-frequency signal of the photovoltaic cell hot spot using the inverse Fourier method. The fourth hot spot identification module generates a second photovoltaic cell hot spot spatial domain according to the hot spot boundary and core region corresponding to the photovoltaic cell hot spot spatial domain based on the temperature change of the high-frequency signal of the photovoltaic cell hot spot. The hot spot localization module uses the optimal adaptive threshold method to perform binarization separation of the hot spot region and background noise in the second photovoltaic cell hot spot spatial domain to obtain the photovoltaic cell hot spot distribution region.

7. A photovoltaic cell stain identification system based on multimodal data fusion according to claim 6, characterized in that, The voltage-current slope model for the first photovoltaic cell is: 。 8. A photovoltaic cell stain identification system based on multimodal data fusion according to claim 6, characterized in that, Second photovoltaic cell voltage-current slope model: ； in: Number of individual photovoltaic panels that are shaded; It blocks the battery's operating current.

Images