Method, system and device for detecting a soiling layer on a photovoltaic module surface based on spectral unmixing

Abstract

The application discloses a photovoltaic module surface contamination layer detection method, system and device based on spectral unmixing. The application uses the spatial-spectral information provided by the hyperspectral image, adopts the initialization spectral unmixing and the robust spectral unmixing with partial end member fixed to obtain the end member spectrum and the corresponding abundance matrix of the existing and potential surface contamination layer, accurately extracts the spatial distribution of the surface contamination layer, and effectively realizes the detection on the surface contamination layer state of the photovoltaic module. The application can continuously monitor, track and warn the power generation efficiency reduction and the potential hot spot risk caused by the contamination layer change, and provides a quantitative basis for the fine decision and optimization of the photovoltaic module cleaning strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical image processing, and more particularly to a photovoltaic module surface contamination layer detection method, system and device based on spectral unmixing. BACKGROUND

[0002] The rapid expansion of the current photovoltaic power generation scale makes the safe operation and operation efficiency optimization of power stations face severe challenges. Among them, the deposition of contaminants on the surface of photovoltaic modules has become a key factor restricting system stability and economy. In the operation process, dust, sand, bird droppings, fallen leaves and other contaminants will gradually accumulate on the surface of the module, forming a shielding contamination layer. This contamination layer not only weakens the light transmission performance of the module, reduces the solar radiation absorption rate, and thus affects the photoelectric conversion efficiency and power output; but also causes uneven distribution and induces local shading (such as point or strip coverage), which induces hot spot effect, and poses a significant threat to the safety and long-term durability of the power station.

[0003] Currently, the monitoring and maintenance of the surface contamination layer of photovoltaic modules still mainly relies on manual experience and qualitative observation, making it difficult to achieve intelligent, high-precision and real-time operation and maintenance requirements. Although cleaning techniques such as pipeline flushing, robot cleaning or vehicle operation have been applied, there is still a lack of data support in terms of cleaning cycle arrangement, path planning and resource management. If only a fixed cycle cleaning strategy is adopted, it is difficult to balance power generation efficiency and operation cost, and it may also cause waste of manpower and resources.

[0004] In recent years, some photovoltaic power stations have begun to introduce unmanned aerial vehicle intelligent inspection technology, which collects component images through visible light and infrared sensors, and realizes pollution layer state detection and analysis through computer vision technology. Although this method can detect hot spots or local shading, it lacks material component analysis capability and cannot accurately quantify the composition and coverage ratio of the pollution layer. Under the influence of diversified pollution layer composition and complex operating environment factors, the detection effect cannot meet the demand of photovoltaic power station for intelligent monitoring and operation and maintenance of the surface pollution layer of photovoltaic components. The hyperspectral image data has the characteristics of extremely high spectral resolution, which gives it the ability to diagnose fine spectral features and identify and analyze ground object types, materials and substance components. The flight platform such as unmanned aerial vehicle carries hyperspectral imager for inspection imaging and real-time analysis and processing, which provides a new effective means for intelligent monitoring of the surface pollution layer of photovoltaic components. In the actual operating environment, the surface pollution layer of photovoltaic components is formed by the adhesion of various liquid and solid particle pollutants in the operating environment on the surface of the components, which is reflected in the hyperspectral image as a mixed spectrum of component material spectrum and various pollution layer component spectrum. The spectral unmixing technology aims to interpret the component spectrum and its composition ratio from the mixed pixel, which provides a new technical approach for analyzing the surface pollution layer state of photovoltaic components. However, the existing spectral unmixing technology applied to the hyperspectral image of photovoltaic components has problems such as spectral variability, serious noise interference and poor robustness, which limits the accuracy of quantitatively solving the surface pollution layer component spectrum and its ratio, and the application effect is poor.

[0005] Therefore, a new scheme is needed to solve the above problems. SUMMARY

[0006] The purpose of the present application is to overcome the shortcomings of the prior art and provide a photovoltaic component surface pollution layer detection method, system and device based on spectral unmixing.

[0007] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0008] The photovoltaic component surface pollution layer detection method based on spectral unmixing comprises:

[0009] Step S1, selecting a single photovoltaic component for cleaning, and collecting hyperspectral images of clean photovoltaic component samples;

[0010] Step S2, selecting a single photovoltaic component with surface pollution layer, and collecting hyperspectral images of surface pollution layer photovoltaic component samples;

[0011] Step S3, collecting hyperspectral images of all photovoltaic components to be detected in the photovoltaic power station by the unmanned aerial vehicle carrying the hyperspectral imager;

[0012] Step S4, preprocessing the collected hyperspectral images;

[0013] Step S5, segmenting and extracting the photovoltaic module region through a photovoltaic module segmentation network;

[0014] Step S6, extracting a photovoltaic module endmember spectrum from the hyperspectral image of the clean photovoltaic module sample;

[0015] Step S7, extracting an existing surface contamination endmember spectrum from the hyperspectral image of the surface contamination photovoltaic module sample;

[0016] Step S8, performing partial endmember fixed initialization spectral unmixing on the hyperspectral images of all photovoltaic modules to be detected in the photovoltaic power station to obtain initial endmember spectra and an abundance matrix;

[0017] Step S9, performing robust spectral unmixing on the hyperspectral image of the photovoltaic module region with surface contamination to obtain the final endmember spectrum and the abundance matrix;

[0018] Step S10, determining the spatial distribution density map and the warning area of the surface contamination of the photovoltaic module.

[0019] Further, in step S5, the photovoltaic module segmentation network adopts one of UNet and Mask R-CNN.

[0020] Further, in step S6, the endmember spectrum extraction method adopts one of the pure pixel index, the simplex growth, and the vertex component analysis, and the extracted photovoltaic module endmember spectrum is where L is the number of wavebands, is the number of photovoltaic module endmembers, is a real number set.

[0021] Further, in step S7, the endmember spectrum extraction method adopts one of the pure pixel index, the simplex growth, and the vertex component analysis, and the extracted existing surface contamination endmember spectrum is where L is the number of wavebands, is the number of contamination endmembers.

[0022] Further, in step S8, the partial endmember fixed initialization spectral unmixing includes the following steps:

[0023] Step S81, setting the number of abnormal surface contamination endmembers The abnormal surface contamination endmember is an endmember of a potential contamination not contained in the surface contamination photovoltaic module sample in step S7;

[0024] Step S82, setting an initial endmember set calculating an initial simplex center

[0025] , ;

[0026] Step S83, calculating the orthogonal projection matrix

[0027]

[0028] ,

[0029] wherein U is the centralized calculation result; I is the unit matrix; the superscript T is the matrix transposition operator; is a regularization parameter;

[0030] Step S84, iteratively extracting the endmember spectrum of the abnormal surface contamination layer ;

[0031] Step S85, based on the endmember spectrum of the photovoltaic module , the endmember spectrum of the surface contamination layer , and the endmember spectrum of the abnormal surface contamination layer to obtain the initial endmember spectrum result , and based on the initial endmember spectrum result , the initial abundance matrix is estimated and solved by full constraint least square method, wherein is the endmember spectrum of the photovoltaic module corresponding to the abundance matrix, is the endmember spectrum of the existing surface contamination layer corresponding to the abundance matrix, is the endmember spectrum of the abnormal surface contamination layer corresponding to the abundance matrix.

[0032] Further, in step S84, the iterative extraction process of the endmember spectrum of the abnormal surface contamination layer is as follows:

[0033] Step S841, for all pixels in the hyperspectral image, the projection distance is calculated, and the pixel with the largest distance is selected as the newly added endmember of the abnormal surface contamination layer :

[0034] ,

[0035]

[0036]

[0037] Step S842, updating the endmember set​ , simplex center

[0038]

[0039]

[0040] Step S843, updating the orthogonal projection matrix

[0041]

[0042]

[0043]

[0044] In the formula, v is a residual vector; w is a projection vector;

[0045] Step S844, repeating the iteration steps S841-S843 for , combining all the new abnormal surface contamination endmember spectra to obtain an abnormal surface contamination endmember spectrum, merging the photovoltaic module endmember spectrum , the surface contamination endmember spectrum and the abnormal surface contamination endmember spectrum to obtain an initial endmember spectrum result .

[0046] Further, in step S9, the calculation method of the robust spectral unmixing is to update the initial endmember spectrum result and the initial abundance matrix

[0047]

[0048]

[0049] In the formula, represents the matrix point multiplication; is a sparse regularization constraint parameter; , is a diagonal matrix, and the lth element on the diagonal is calculated as

[0050]

[0051] .

[0052] Further, in step S10, the spatial distribution density map and the warning area of the surface contamination of the photovoltaic module are determined, including the following steps:

[0053] Step S101, from the initial abundance matrix The abundance matrix corresponding to the surface contaminant endmembers is obtained by disassembly. , ;

[0054] Step S102: For each pixel j, calculate the total abundance of surface contaminants.

[0055] ,

[0056] Step S103: Calculate the total abundance of surface contaminants for each pixel. Mapped to a two-dimensional matrix based on pixel coordinates. ,in and These are the height and width of the image, respectively.

[0057] Step S104, when the total abundance of surface contaminants on the pixel Greater than the dynamic threshold When this happens, the pixel is designated as a warning zone, and the dynamic threshold is applied. The calculation method is as follows

[0058]

[0059] In the formula, Basic threshold; This is the adjustment factor; PM is the air quality index for the day. and Historical air quality index mean and standard deviation.

[0060] The present invention also provides a photovoltaic module surface contamination detection system based on spectral unmixing, the system comprising:

[0061] The hyperspectral image acquisition unit is used to acquire hyperspectral images of the photovoltaic module under test.

[0062] The preprocessing unit is used to perform reflectance correction and outlier removal on the hyperspectral image.

[0063] Component segmentation unit, used to segment and extract areas of photovoltaic modules;

[0064] An end-member extraction unit is used to extract the end-member spectra of photovoltaic modules and the end-member spectra of existing surface contaminants.

[0065] The spectral unmixing unit is used for initial spectral unmixing with partially fixed endmembers and robust spectral unmixing to obtain endmember spectra and abundance matrices.

[0066] A surface contamination detection unit is configured to determine a spatial distribution density map of the surface contamination of the photovoltaic module and a warning area.

[0067] The application also provides a device for detecting surface contamination of a photovoltaic module based on spectral unmixing, comprising:

[0068] A memory is configured to store computer programs and data.

[0069] A processor is configured to implement the steps of the method for detecting surface contamination of a photovoltaic module based on spectral unmixing when executing the computer programs.

[0070] The application has the following advantages:

[0071] The application quantitatively analyzes the spatial distribution of the surface contamination of a photovoltaic module based on spectral unmixing technology, and can robustly and accurately detect the state of the contamination. The technology can continuously monitor, track and warn about the decrease in power generation efficiency and the potential hot spot risk caused by the contamination, and provide quantitative basis for fine decision-making and optimization of the cleaning strategy of the photovoltaic module. This not only helps to ensure the safe operation of the power station and save cleaning resources, but also has application value for improving the operation and maintenance efficiency of the photovoltaic power station. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A flowchart of the method for detecting surface contamination of a photovoltaic module based on spectral unmixing in the embodiment;

[0073] Figure 2 A structural framework diagram of the system for detecting surface contamination of a photovoltaic module based on spectral unmixing in the embodiment;

[0074] Figure 3 A structural framework diagram of the device for detecting surface contamination of a photovoltaic module based on spectral unmixing in the embodiment.

[0075] Reference signs: hyperspectral image acquisition unit 1, preprocessing unit 2, module segmentation unit 3, endmember extraction unit 4, spectral unmixing unit 5, surface contamination detection unit 6, memory 7, processor 8. DETAILED DESCRIPTION

[0076] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0077] Embodiment: a method for detecting surface contamination of a photovoltaic module based on spectral unmixing, as shown in the figure, the method comprises: Figure 1 ​

[0078] Step S1: Select a single photovoltaic module for cleaning and acquire a hyperspectral image of the cleaned photovoltaic module sample.

[0079] Step S2: Select a single photovoltaic module with a surface contamination layer and acquire a hyperspectral image of the photovoltaic module sample with the surface contamination layer.

[0080] Step S3: Use a drone equipped with a hyperspectral imager to collect hyperspectral images of all photovoltaic modules to be tested in the photovoltaic power station.

[0081] Step S4: Preprocess the acquired hyperspectral image, including reflectance correction and outlier removal;

[0082] The method for calculating reflectivity correction is as follows:

[0083]

[0084] In the formula, To be at wavelength Reflectance at that location; To be at wavelength The radiance at that location (which can be obtained from the sensor); The dark current response was obtained by measuring the black calibration plate; The radiance obtained by measuring a white reference plate; The known reflectance value of the white reference plate, typically 1 or a known high reflectance value.

[0085] The outlier removal method is as follows:

[0086]

[0087] In the formula, To be at wavelength The reflectivity at that location.

[0088] Step S5: The photovoltaic module region is segmented and extracted using a photovoltaic module segmentation network; the photovoltaic module segmentation network is either UNet or Mask R-CNN.

[0089] Step S6: Extract the end-member spectrum of the photovoltaic module from the hyperspectral image of the clean photovoltaic module sample;

[0090] Among them, the endmember spectral extraction method employs one of the following: pure pixel index, simplex growth, and vertex component analysis. The extracted photovoltaic module endmember spectrum is... Where L is the number of bands, The number of terminal elements in a photovoltaic module. It is the set of real numbers.

[0091] Step S7, extracting existing surface contamination endmember spectrum from hyperspectral image of surface contamination photovoltaic module sample;

[0092] Wherein, the endmember spectrum extraction method adopts one of pure pixel index, simplex growing and vertex component analysis, and the extracted existing surface contamination endmember spectrum is Wherein, L is the number of wavebands, is the number of contamination endmembers.

[0093] Step S8, performing partial endmember fixed initialization spectrum unmixing on hyperspectral image of all photovoltaic modules to be detected in photovoltaic power station to obtain initial endmember spectrum and abundance matrix;

[0094] Specifically, the method comprises the following steps:

[0095] Step S81, setting abnormal surface contamination endmember number The abnormal surface contamination endmember is an endmember of potential contamination not contained in the surface contamination photovoltaic module sample in step S7; Usually set to 1-3;

[0096] Step S82, setting initial endmember set Calculating initial simplex center

[0097] , ;

[0098] Step S83, calculating orthogonal projection matrix

[0099]

[0100] ,

[0101] Wherein, U is centered calculation result; I is unit matrix; superscript T is matrix transposition operator; is regularization parameter;

[0102] Step S84, for iteratively extracting abnormal surface contamination endmember spectrum ; Specifically:

[0103] Step S841, for all pixels in the hyperspectral image, calculating projection distance , and selecting the pixel with the largest distance as new abnormal surface contamination endmember :

[0104] ,

[0105]

[0106]

[0107] Step S842, Update the end-user set Simplex center

[0108]

[0109]

[0110] Step S843, update the orthogonal projection matrix

[0111]

[0112]

[0113]

[0114] In the formula, v is the residual vector; w is the projection vector;

[0115] Step S844, for Repeat steps S841-S843 to add all newly added abnormal contamination endmembers. Combining these spectra yields the end-member spectra of the abnormal surface contaminant layer, and the combined end-member spectra of the photovoltaic module are then obtained. Surface contaminant end-member spectrum and the end-member spectrum of abnormal surface contaminants The initial endmember spectral results were obtained. ;

[0116] Step S85, based on the initial endmember spectral results Based on this, the initial abundance matrix is ​​obtained by estimating and solving using the fully constrained least squares method. ,in It is the end-member spectrum of photovoltaic modules The corresponding abundance matrix, It is the existing surface contaminant end-member spectrum The corresponding abundance matrix, It is the end-member spectrum of the abnormal surface contaminant layer. The corresponding abundance matrix.

[0117] Step S9: Robustly unmix the hyperspectral image of the photovoltaic module region with surface contamination to obtain the final endmember spectrum and abundance matrix.

[0118] The calculation method of the robust spectral unmixing is to repeatedly update the initial endmember spectrum result through the following calculation formula and the initial abundance matrix

[0119]

[0120]

[0121] In the formula, denotes matrix point multiplication; is a sparse regularization constraint parameter; , is a diagonal matrix, and the lth element on the diagonal is calculated as

[0122]

[0123] .

[0124] Step S10, determining the spatial distribution density map of the surface contamination layer of the photovoltaic module and the warning area;

[0125] Specifically, the following steps are included:

[0126] Step S101, decomposing and obtaining the abundance matrix of the surface contamination layer endmember from the initial abundance matrix .

[0127] Step S102, calculating the total abundance of the surface contamination layer of each pixel j

[0128] ,

[0129] Step S103, mapping the total abundance of the surface contamination layer of each pixel to a two-dimensional matrix according to the pixel coordinates, wherein and are the height and width of the image respectively;

[0130] Step S104, when the total abundance of the surface contamination layer of the pixel is greater than the dynamic threshold , the pixel is classified as a warning area, and the dynamic threshold is calculated as

[0131]

[0132] In the formula, is a basic threshold, which can be set as [0.1, 0.3] generally.​​ is the coefficient of adjustment; PM is the air index of the day, generally PM10 index can be taken, and the historical air index mean and standard deviation.

[0133] In the technical scheme provided by the embodiment, first, a hyperspectral image of a clean photovoltaic module sample is collected; then, a hyperspectral image of a surface contamination photovoltaic module sample is collected; and then, a hyperspectral image of all photovoltaic modules to be detected is collected; the hyperspectral image is preprocessed, including reflectivity correction and abnormal value elimination; a photovoltaic module endmember spectrum is extracted from the hyperspectral image of the clean photovoltaic module sample; an existing surface contamination endmember spectrum is extracted from the hyperspectral image of the surface contamination photovoltaic module sample; then, the hyperspectral image of all photovoltaic modules to be detected in the photovoltaic power station is subjected to partial endmember fixed initialization spectral unmixing to obtain an initial endmember spectrum and an abundance matrix; and then, the hyperspectral image of the region of the photovoltaic module with surface contamination is subjected to robust spectral unmixing to obtain a final endmember spectrum and an abundance matrix; and the spatial distribution density map of the surface contamination of the photovoltaic module and the early warning region are determined.

[0134] In the technical scheme provided by the embodiment, the spatial-spectral information provided by the hyperspectral image is used to obtain the endmember spectrum and the corresponding abundance matrix of the existing and potential surface contamination through partial endmember fixed initialization spectral unmixing and robust spectral unmixing, and the spatial distribution of the surface contamination is accurately extracted, so that the detection of the state of the surface contamination of the photovoltaic module is effectively realized.

[0135] See Figure 2 , a structure diagram of a photovoltaic module surface contamination detection system based on spectral unmixing provided by the embodiment, the system comprising a hyperspectral image acquisition unit 1, a preprocessing unit 2, a component segmentation endmember, an endmember extraction unit 4, a spectral unmixing unit 5, and a surface contamination detection unit 6.

[0136] The hyperspectral image acquisition unit 1 is configured to collect data of the photovoltaic module to be detected to obtain a hyperspectral image.

[0137] The preprocessing unit 2 is configured to perform reflectivity correction and abnormal value elimination preprocessing on the hyperspectral image.

[0138] The component segmentation unit 3 is configured to segment and extract the photovoltaic module region.

[0139] The endmember extraction unit 4 is configured to extract a photovoltaic module endmember spectrum and an existing surface contamination endmember spectrum.

[0140] The spectral unmixing unit 5 is configured to perform partial endmember fixed initialization spectral unmixing and robust spectral unmixing to obtain an endmember spectrum and an abundance matrix.

[0141] The surface contamination detection unit 6 is used to determine the spatial distribution density map of the surface contamination of the photovoltaic module and the early warning area.

[0142] In the technical scheme provided by the embodiment,

[0143] First, a single photovoltaic module is selected for cleaning, and a hyperspectral image of the clean photovoltaic module sample is collected; then, a single photovoltaic module with surface contamination is selected, and a hyperspectral image of the surface contamination photovoltaic module sample is collected; then, a hyperspectral imager is carried by a drone to collect hyperspectral images of all photovoltaic modules to be detected in the photovoltaic power station; then, the collected hyperspectral images are preprocessed, including reflectance correction and outlier rejection; then, the photovoltaic module region is segmented and extracted by a photovoltaic module segmentation network; then, the endmember spectrum of the photovoltaic module is extracted from the hyperspectral image of the clean photovoltaic module sample; then, the existing surface contamination endmember spectrum is extracted from the hyperspectral image of the surface contamination photovoltaic module sample; then, the hyperspectral images of all photovoltaic modules to be detected in the photovoltaic power station are partially endmember fixed for initialization spectral unmixing, to obtain the initial endmember spectrum and abundance matrix; then, the hyperspectral image of the photovoltaic module region with surface contamination is robustly spectrally unmixed to obtain the final endmember spectrum and abundance matrix; finally, the spatial distribution density map of the surface contamination of the photovoltaic module and the early warning area are determined.

[0144] See Figure 3 A structural schematic diagram of a photovoltaic module surface contamination detection device based on spectral unmixing is provided in the embodiment. The device includes a memory 7 and a processor 8 connected to the memory 7.

[0145] The memory 7 is used to store computer programs and data; the processor 8 is used to execute the computer programs to realize the steps of the above-mentioned photovoltaic module surface contamination detection method based on spectral unmixing.

[0146] The above-mentioned only is the preferred embodiment of the present application, the protection scope of the present application is not only limited to the above-mentioned embodiment, all technical solutions belonging to the idea of the present application are within the protection scope of the present application. It should be pointed out that, for ordinary skilled in the art, some improvements and decorations without departing from the principle of the present application, these improvements and decorations should also be considered as the protection scope of the present application.

Claims

1. A method for detecting a surface soiling layer of a photovoltaic module based on spectral unmixing, characterized in that, The method comprises: Step S1, selecting a single photovoltaic module for cleaning, and collecting a hyperspectral image of a clean photovoltaic module sample; Step S2, selecting a single photovoltaic module with a surface contamination layer, and collecting a hyperspectral image of a surface contamination photovoltaic module sample; Step S3, collecting hyperspectral images of all photovoltaic modules to be detected in the photovoltaic power station by a drone carrying a hyperspectral imager; Step S4, preprocessing the collected hyperspectral images; Step S5, segmenting and extracting the photovoltaic module region by a photovoltaic module segmentation network; Step S6, extracting the photovoltaic module endmember spectrum from the hyperspectral image of the clean photovoltaic module sample; Step S7, extracting the existing surface contamination endmember spectrum from the hyperspectral image of the surface contamination photovoltaic module sample; Step S8, performing partial endmember fixed initialization spectral unmixing on the hyperspectral images of all photovoltaic modules to be detected in the photovoltaic power station to obtain initial endmember spectra and abundance matrixes; Step S9, performing robust spectral unmixing on the hyperspectral images of the photovoltaic module region with the surface contamination layer to obtain final endmember spectra and abundance matrixes; Step S10, determining the spatial distribution density map and warning area of the photovoltaic module surface contamination layer; In step S8, the partial endmember fixed initialization spectral unmixing comprises the following steps: Step S81, set abnormal surface contamination end-member number The abnormal surface contamination end-member is an end-member of potential contamination not contained in the surface contamination photovoltaic module sample in step S7. Step S82, set initial endmember set , calculate initial simplex centers ， ； Step S83, calculate the orthogonal projection matrix ， where U is centralized computation result; I is an identity matrix; the superscript T is a matrix transposition operator; is a regularization parameter; Step S84, for , iteratively extract the endmember spectrum of the abnormal surface contamination layer ; Step S85, based on the end-member spectrum of the photovoltaic module Surface contaminant end-member spectrum and the end-member spectrum of abnormal surface contaminants To obtain initial endmember spectral results Based on the initial endmember spectral results The initial abundance matrix is ​​obtained by estimating and solving using the fully constrained least squares method. ,in It is the end-member spectrum of photovoltaic modules The corresponding abundance matrix, It is the existing surface contaminant end-member spectrum The corresponding abundance matrix, It is the end-member spectrum of the abnormal surface contaminant layer. The corresponding abundance matrix; In step S9, the calculation method of robust spectral unmixing is to update the initial endmember spectrum results by repeating iteration through the following calculation formula and the initial abundance matrix wherein denotes a matrix point multiplication; is a sparse regularization constraint parameter; , is a diagonal matrix whose lth diagonal element is calculated as 。 2. The method of claim 1, wherein the method is based on spectral unmixing. In step S5, the photovoltaic module segmentation network uses one of UNet and Mask R-CNN.

3. The method of claim 1, wherein the method is based on spectral unmixing. In step S6, the end-member spectrum extraction method adopts one of the pure pixel index, simplex growth and vertex component analysis, and the extracted photovoltaic module end-member spectrum is where L is the number of wave bands, is the number of photovoltaic module end-members, is a real number set.

4. The method of claim 1, wherein the method is based on spectral unmixing. In step S7, the endmember spectrum extraction method adopts one of the pure pixel index, simplex growth and vertex component analysis, and the extracted existing surface contamination endmember spectrum is wherein L is the number of wavebands, is the number of contamination endmembers.

5. The method of claim 1, wherein the method is based on spectral unmixing. In step S84, the end-member spectrum of the abnormal surface contaminant layer is obtained. The iterative extraction process is as follows: Step S841, for all pixels in the hyperspectral image , the projection distance is calculated , and the pixel with the largest distance is selected as the new abnormal surface contamination endmember ， Step S842, update the endomembrane set , simplex center Step S843, updating the orthogonal projection matrix In the formula, v is a residual vector; w is a projection vector; Step S844, for , repeat iteration steps S841-S843, combine all new abnormal soiling layer endmember spectra together to get abnormal surface soiling endmember spectrum, combine photovoltaic module endmember spectrum , surface soiling endmember spectrum , and abnormal surface soiling endmember spectrum to get initial endmember spectrum result .

6. The method of claim 1, wherein the method is based on spectral unmixing. In step S10, determining the spatial distribution density map and warning area of the photovoltaic module surface contamination layer comprises the following steps: Step S101, the initial abundance matrix is decomposed to obtain the abundance matrix corresponding to the surface contamination endmember , ;​ Step S102, for each pixel j, calculate its total surface contamination abundance ， Step S103, total abundance of surface contamination of each pixel is mapped to a two-dimensional matrix mapping to a two-dimensional matrix wherein and are the height and width of the image, respectively Step S104, when the total abundance of surface contamination of the pixel is greater than a dynamic threshold The pixel is classified as a pre-warning area The dynamic threshold is calculated as wherein is a base threshold value; is an adjustment factor; PM is the daily air pollution index, and the historical air pollution index mean and standard deviation.

7. A spectral unmixing based photovoltaic module surface soiling detection system for implementing the method of claim 1, characterized in that, The system comprises: A hyperspectral image acquisition unit (1) for collecting data of the photovoltaic module to be detected to obtain a hyperspectral image; A preprocessing unit (2) for performing reflectivity correction and outlier rejection preprocessing on the hyperspectral image; A module segmentation unit (3) for segmenting and extracting the photovoltaic module region; An endmember extraction unit (4) for extracting the photovoltaic module endmember spectrum and the existing surface contamination endmember spectrum; A spectral unmixing unit (5) for partial endmember fixed initialization spectral unmixing and robust spectral unmixing to obtain the endmember spectrum and the abundance matrix; A surface contamination detection unit (6) for determining the spatial distribution density map and warning area of the photovoltaic module surface contamination layer.

8. A device for detecting a soiling layer on a photovoltaic module surface based on spectral unmixing, characterized by Comprise: A memory (7) for storing computer programs and data; A processor (8) for executing computer programs to realize the steps of the spectral unmixing-based photovoltaic module surface contamination detection method according to any one of claims 1-6.

Images