Photovoltaic system efficiency evaluation method and system based on unmanned aerial vehicle multi-source image fusion

Abstract

The invention relates to the field of new energy, and discloses a photovoltaic system efficiency evaluation method and system based on unmanned aerial vehicle multi-source image fusion, and the method comprises the steps: collecting a visible light orthoimage, thermal imaging data and environment parameters of a photovoltaic system, and obtaining a registration result through employing an improved feature point matching algorithm; based on a registration result, combining spatial features of a visible light image and temperature features of thermal imaging, realizing accurate segmentation of the photovoltaic panel through a deep learning model, and identifying the type, the arrangement mode and the installation angle of the photovoltaic panel at the same time; establishing a mathematical model of the relation between the photovoltaic panel temperature distribution and the power generation efficiency, distinguishing the normal working temperature difference and the fault hot spot, and analyzing and determining the efficiency attenuation degree and the fault type of the photovoltaic system through a heat distribution mode. The method is accurate in image registration, good in data fusion effect, accurate in temperature anomaly detection and comprehensive in efficiency evaluation, and provides more efficient and accurate technical support for management and maintenance of the distributed photovoltaic system.

Description

Technical Field

[0001] The present invention relates to the field of new energy technologies, and in particular to a photovoltaic system efficiency evaluation method and system based on unmanned aerial vehicle (UAV) multi-source image fusion. Background Art

[0002] Photovoltaic power generation systems, as a vital component of clean energy, are increasingly being used on urban rooftops. Existing PV system efficiency assessment technologies primarily rely on manual on-site inspections, fixed monitoring systems, or single-image detection methods. Manual inspections utilize devices such as infrared thermometers and multimeters, while fixed monitoring systems record electrical parameters through data loggers. Single-image detection utilizes visible light cameras or thermal imaging cameras for capture and analysis. Each of these methods has its own technical limitations. Visible light images cannot detect hidden thermal effects, while thermal imaging struggles to accurately identify the location and boundaries of PV panels. Furthermore, they lack effective image registration and data fusion algorithms, resulting in incomplete and inaccurate assessment results.

[0003] Existing drone image detection technology primarily uses a single sensor to collect data. This approach typically involves capturing images first, followed by manual analysis or simple algorithmic analysis. This approach suffers from significant flaws in image processing: First, there is a lack of effective multi-source image registration technology, making it difficult to accurately align thermal and orthophoto images. Second, there is no multi-source data fusion algorithm tailored to photovoltaic panel characteristics, making it impossible to simultaneously utilize both thermal image temperature information and orthophoto spatial information. Third, there is a lack of mechanisms to address interfering factors such as shadows and reflections in rooftop environments, reducing recognition accuracy. Fourth, the evaluation model is overly simplified and cannot accurately reflect the actual operating efficiency of the photovoltaic system.

[0004] Existing methods for detecting the temperature characteristics of photovoltaic systems suffer from significant technical flaws. First, thermal imaging temperature measurements fail to account for the influence of ambient temperature, roof material, and solar radiation intensity, resulting in low temperature measurement accuracy. Second, the automatic detection algorithm for temperature anomalies is simplistic and cannot accurately distinguish between normal temperature differences and fault hotspots. Third, the correlation model between temperature data and power generation efficiency is overly simplified, failing to accurately quantify the impact of temperature on efficiency. Fourth, there is a lack of compensation mechanisms to account for the impact of photovoltaic panel material and installation angle on thermal imaging results, reducing the accuracy of the assessment. Summary of the Invention

[0005] In response to the above-mentioned technical deficiencies, the technical problem to be solved by the present invention is to provide a photovoltaic system efficiency evaluation method and system based on drone multi-source image fusion, aiming to solve the technical problems existing in the existing technology in the efficiency evaluation of rooftop photovoltaic systems, such as inaccurate image registration, poor data fusion effect, inaccurate temperature anomaly detection and incomplete efficiency evaluation.

[0006] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a photovoltaic system efficiency evaluation method based on UAV multi-source image fusion, comprising: S1. Collect visible light orthophotos, thermal imaging data and environmental parameters of the photovoltaic system; S2. Based on the collected visible light orthophotos, thermal imaging data, and environmental parameters, an improved feature point matching algorithm is used to obtain a high-precision spatial registration result between the thermal image and the orthophoto. S3. Based on the registration results and the spatial characteristics of the visible light image and the temperature characteristics of the thermal image, the deep learning model is used to automatically identify the position of the photovoltaic panels, achieve accurate segmentation of the photovoltaic panels, and simultaneously identify the characteristic information of the photovoltaic panel type, arrangement, and installation angle; S4. Establish a mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency, intelligently identify and classify temperature anomalies, distinguish between normal operating temperature differences and fault hotspots, and determine the degree of photovoltaic system efficiency attenuation and potential fault types through thermal distribution pattern analysis.

[0007] Furthermore, the step S2 includes: S21, performing image stitching processing on the collected visible light orthophotos to generate orthophotos; S22. Perform geometric correction and radiation correction on thermal imaging; S23. Combining the edge and corner features of the photovoltaic panel, the ORB algorithm is used to extract feature points, and preliminary matching is performed using the Hamming distance and FLANN fast approximate nearest neighbor method. The RANSAC algorithm is used to eliminate erroneous points and calculate the homography matrix and the basic matrix. Then, based on the improved NCC method, bilinear interpolation and bidirectional verification mechanism are combined to adapt to rotation and scale changes and correct deformations caused by shooting angles and sensor differences. Finally, by dynamically adjusting the search window size, combining the RMSE elimination mechanism and the correlation coefficient threshold to screen matching points, the registration accuracy is better than 2 pixels.

[0008] Furthermore, step S3 includes: S31. Extract texture and geometric features from visible light orthophotos; S32, combining temperature information from thermal imaging to assist boundary optimization; S33, fusing visible light spatial features with thermal imaging temperature features through a multi-source feature fusion formula; S34. Inputting the fused feature vector into a deep learning model to accurately segment the photovoltaic panels and simultaneously identify feature information including the type, arrangement, and installation angle of the photovoltaic panels; S35. Post-process the recognition results to eliminate noise and misidentification, and generate a photovoltaic panel area mask.

[0009] Furthermore, the multi-source feature fusion formula is as follows: Fusion = α × FRGB + β × FIR + c × Fgeo in, Fusion is the fused feature vector, FRGB is the spatial feature vector extracted from the visible light image, FIR The temperature feature vector extracted from thermal imaging, Fgeo is the geometric shape eigenvector, α 、 β 、 c is the weight coefficient, and α + β + c =1.

[0010] Furthermore, the deep learning model loss function is as follows: L total =L seg +λ 1 L cls +λ 2 L reg L total =L seg +λ 1 L cls +λ 2 L reg in, L total is the total loss function, L seg is the segmentation loss, L cls is the classification loss, used for photovoltaic panel type recognition, L reg is the regression loss, used for geometric parameter prediction, l 1 and l 2 is the loss weight balance parameter; The Dice Loss segmentation loss is calculated as follows: L seg =1−2|P∩G|||P|+|G| L seg =1−| P ∣+∣ G ∣2∣P ∩ G ∣ in, P is the set of photovoltaic panel pixels predicted by the model, G is the set of real labeled photovoltaic panel pixels, | P ∩ G ∣ is the number of intersection pixels between the prediction and the true annotation, ∣ P ∣ is the total number of predicted pixels, ∣ G ∣ is the total number of real labeled pixels; The confidence level of photovoltaic panel identification is calculated as follows:

[0011] in, Cpanel is the confidence level of photovoltaic panel identification, Stem is the temperature feature score, Shape is the shape feature score, Texture is the texture feature score, w 1. w 2 and w 3 is the feature weight parameter, b is the bias parameter, e is the base of natural logarithm; The photovoltaic panel installation angle is calculated as follows:

[0012] in, i The installation tilt angle of the photovoltaic panel, H max is the highest point height of the photovoltaic panel, H min is the lowest point height of the photovoltaic panel, L panel is the length of the photovoltaic panel, and arctan is the inverse tangent function.

[0013] Furthermore, the step S4 includes: S41, establishing a reference temperature area by measuring the normal operating temperature of surrounding non-faulty photovoltaic panels as a benchmark; S42. Analyze the surface temperature distribution of photovoltaic panels through thermal imaging and identify areas with abnormal temperature; S43, using the set temperature anomaly detection threshold to classify the temperature anomaly points into three levels: slight anomaly, moderate anomaly, and severe anomaly; S44. Determine the type of abnormality based on the distribution pattern of the abnormal temperature points; S45. Evaluate the efficiency loss of the photovoltaic system based on the degree and distribution characteristics of temperature anomalies.

[0014] Furthermore, the temperature anomaly detection threshold formula is as follows:

[0015] in, T threshold is the temperature anomaly detection threshold, is the average temperature of the reference area, sref is the standard deviation of the reference area temperature, k is the threshold coefficient.

[0016] Furthermore, the mathematical model of the relationship between the photovoltaic panel temperature distribution and power generation efficiency is as follows: η = η0× [1 - β × (T - T0) - γ × (Nh / Nt) - δ × (ΔT / T0)] Where η is the actual efficiency, η0 is the efficiency under standard conditions, β is the temperature coefficient, T is the average temperature of the photovoltaic panel, T0 is the reference temperature, Nh is the number of hot spots, Nt is the total number of photovoltaic panels, ΔT is the difference between the highest temperature and the average temperature, and γ and δ are coefficients related to equipment aging and uneven heat distribution.

[0017] Furthermore, the mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency also includes: When applied to the efficiency evaluation of large-scale ground-based photovoltaic power plants, the mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency is adjusted according to the characteristics of the ground-based power plants, and parameters for dust coverage and local shading factors are added: η = η0 × [1 - β × (T - T0) - γ × (Nh / Nt) - δ × (ΔT / T0) - ε × D] Where D is the dust cover index, which is calculated by the reflectivity change of visible light images, and ε is the dust influence coefficient, which is calibrated according to the environmental conditions in different regions.

[0018] Photovoltaic system efficiency evaluation system based on UAV multi-source image fusion, including: It consists of three parts: data acquisition subsystem, data processing subsystem and efficiency evaluation subsystem; The data acquisition subsystem includes a UAV platform, a visible light camera, a thermal imaging camera and an environmental parameter acquisition module, which is used to collect visible light orthophotos, thermal imaging data and environmental parameters of the photovoltaic system; Among them, the environmental parameter acquisition module includes a temperature sensor, a light intensity sensor and a wind speed sensor; The data processing subsystem includes a data transmission module, a ground control station, an image preprocessing unit, a multi-source image registration module, and a photovoltaic panel identification and segmentation module. It uses an improved feature point matching algorithm to obtain high-precision spatial registration results. Combining the spatial characteristics of visible light images with the temperature characteristics of thermal imaging, it uses a deep learning model to accurately segment photovoltaic panels and simultaneously identify characteristic information such as panel type, arrangement, and installation angle. The efficiency evaluation subsystem includes a temperature anomaly detection module, an efficiency evaluation module, and a result output module. It is used to establish a mathematical model of the relationship between the temperature distribution of photovoltaic panels and power generation efficiency, intelligently identify and classify temperature anomalies, distinguish between normal operating temperature differences and fault hotspots, and determine the degree of efficiency attenuation and potential fault types of the photovoltaic system through thermal distribution pattern analysis.

[0019] The beneficial effects of the present invention are: combining drone technology, multi-source image processing and artificial intelligence analysis, compared with the existing technology, firstly, it improves the accuracy of efficiency evaluation, comprehensively reflects the working status of the photovoltaic system through multi-source data fusion, and improves the evaluation accuracy by more than 30%; secondly, it realizes fully automated analysis, and can complete the entire process from data collection to efficiency evaluation without human intervention, and the evaluation efficiency is improved by more than 5 times; thirdly, the system has strong adaptability to the environment and can accurately evaluate photovoltaic systems with different types of roofs and different installation methods; fourthly, it provides standardized evaluation processes and results, which can be used for efficiency monitoring and maintenance management of large-scale distributed photovoltaic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is a diagram of the improved U-Net3+ network structure provided by an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the improved U-Net3+ precise segmentation method of the present invention.

[0023] Figure 3 Schematic diagram of image classification based on feature heat in the present invention.

[0024] Figure 4 Schematic diagram of thermal analysis of rooftops in different areas according to the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example

[0026] like Figure 1~Figure 4 As shown, this embodiment uses a drone equipped with both a visible light camera and a thermal imaging camera to collect multi-source data from a rooftop photovoltaic system, and uses innovative data processing algorithms to achieve multi-source image fusion and analysis. A photovoltaic system efficiency evaluation method based on drone multi-source image fusion is provided, including: S1. Collect visible light orthophotos, thermal imaging data, and environmental parameters of the photovoltaic system. The following should be specified: The drone is equipped with a high-resolution visible light camera and an infrared thermal imaging camera. By designing a specific flight path and shooting parameters, it can simultaneously obtain visible light orthophotos and thermal imaging data of the photovoltaic system. This embodiment uses GPS or RTK positioning technology to ensure the accuracy of the geographic reference of the image, and simultaneously records environmental parameters such as ambient temperature and solar radiation intensity to provide a correction basis for subsequent data processing.

[0027] S2. Based on the collected visible light orthophotos, thermal imaging data, and environmental parameters, an improved feature point matching algorithm is used to perform multi-level feature extraction, a non-rigid transformation model, and sub-pixel precision registration optimization based on the differences in thermal and orthophoto characteristics, resulting in a high-precision spatial registration result between the thermal and orthophotos. It should be noted that: S21, performing image stitching processing on the collected visible light orthophotos to generate orthophotos; S22. Perform geometric correction and radiation correction on thermal imaging; S23. Based on the edge and corner features of photovoltaic panels, the ORB algorithm is used to extract feature points. Initial matching is performed using the Hamming distance and FLANN fast approximate nearest neighbor method. False points are removed using the RANSAC algorithm, and the homography matrix and fundamental matrix are calculated. Then, based on an improved NCC method, combined with bilinear interpolation and a bidirectional verification mechanism, the initial matching only utilizes local information of feature points, rather than the geometric relationship between the image pairs. Therefore, geometric constraints are first used to select candidate points, and then an improved NCC (M-NCC) method is used for feature matching. Traditional NCC methods based on rectangular windows are not robust to rotation and scale changes. However, visual images from mobile platforms often experience rotation and scale changes. Therefore, we modified the traditional NCC method by distorting the correlation window to make it robust in the presence of such geometric distortions. Specifically, a rectangular window is opened in the reference image and projected onto the input image using the homomorphic graph obtained in the initial matching, which may be an irregular quadrilateral. The irregular quadrilateral is then resampled to the same size as the rectangular window using bilinear interpolation. Finally, the correlation coefficient of the rectangular window is calculated according to the standard NCC. The resulting transformation matrix uses the RMSE to remove all possible erroneous pairs of homonymous points. If the RMSE is greater than 2 pixels, the homonymous points are treated as erroneous pairs and deleted.

[0028] S3, based on the registration results, combined with the spatial features of the visible light image and the temperature features of the thermal imaging, Figure 1 The improved U-Net3+ network structure shown automatically identifies the location of photovoltaic panels, achieves accurate segmentation of photovoltaic panels, and simultaneously identifies the characteristic information of photovoltaic panel type, arrangement and installation angle. S31. Extract texture and geometric features from visible light orthophotos. S32, combining temperature information from thermal imaging to assist boundary optimization; S33, fusing visible light spatial features with thermal imaging temperature features through a multi-source feature fusion formula; The multi-source feature fusion formula is as follows: Fusion = α × FRGB + β × FIR + c × Fgeo in, Fusion is the fused feature vector, FRGB is the spatial feature vector extracted from the visible light image, FIR The temperature feature vector extracted from thermal imaging, Fgeo is the geometric shape eigenvector, α 、β 、 c is the weight coefficient, and α + β + c =1.

[0029] S34. Inputting the fused feature vector into a deep learning model to accurately segment the photovoltaic panels and simultaneously identify feature information including the type, arrangement, and installation angle of the photovoltaic panels; The deep learning model loss function is as follows: L total =L seg +λ 1 L cls +λ 2 L reg L total =L seg +λ 1 L cls +λ 2 L reg in, L total is the total loss function, L seg is the segmentation loss, L cls is the classification loss, used for photovoltaic panel type recognition, L reg is the regression loss, used for geometric parameter prediction, l 1 and l 2 is the loss weight balance parameter; The Dice Loss segmentation loss is calculated as follows: Lseg=1−2∣P∩G∣∣P∣+∣G∣ Lseg =1−| P ∣+∣ G ∣2∣ P ∩ G ∣ in, P is the set of photovoltaic panel pixels predicted by the model, G is the set of real labeled photovoltaic panel pixels, | P ∩ G ∣ is the number of intersection pixels between the prediction and the true annotation, ∣ P ∣ is the total number of predicted pixels, ∣ G ∣ is the total number of real labeled pixels; The confidence level of photovoltaic panel identification is calculated as follows:

[0030] in, C panel The confidence level of photovoltaic panel identification is between 0 and 1. S temp is the temperature feature score, S shape is the shape feature score, S texture is the texture feature score, w 1. w 2 and w 3 is the feature weight parameter, b is the bias parameter, e is the base of natural logarithm; The photovoltaic panel installation angle is calculated as follows:

[0031] in, i The installation tilt angle of the photovoltaic panel, H max is the highest point height of the photovoltaic panel, H min is the lowest point height of the photovoltaic panel, L panel is the length of the photovoltaic panel, and arctan is the inverse tangent function.

[0032] S35, such as Figure 2 As shown, the recognition results are post-processed to eliminate noise and misidentification and generate a photovoltaic panel area mask.

[0033] S4. Establish a mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency, intelligently identify and classify temperature anomalies, distinguish between normal operating temperature differences and fault hotspots, and determine the degree of photovoltaic system efficiency degradation and potential fault types through thermal distribution pattern analysis. S41, establishing a reference temperature area by measuring the normal operating temperature of surrounding non-faulty photovoltaic panels as a benchmark; S42. Analyze the surface temperature distribution of photovoltaic panels through thermal imaging and identify areas with abnormal temperature; S43, such as Figure 3 and Figure 4 As shown, Figure 3SolarPanel 0.93 and Solar Panel 0.91 represent solar panels and their efficiency values. For example, SolarPanel 0.93 represents a solar panel with an efficiency of 0.93. In this step, the set temperature anomaly detection threshold (generally the reference temperature +10°C) is used to classify temperature anomalies into three levels: mild anomaly (reference temperature +10°C to +20°C), moderate anomaly (reference temperature +20°C to +30°C), and severe anomaly (reference temperature above +30°C). The temperature anomaly detection threshold formula is as follows:

[0034] in, T threshold is the temperature anomaly detection threshold, is the average temperature of the reference area, sref is the standard deviation of the reference area temperature, k is the threshold coefficient, which is usually 2~3.

[0035] S44. Determine the type of anomaly (e.g., hot spot, connection failure, local shadow, etc.) based on the distribution pattern of the temperature anomaly points. S45. Evaluate the efficiency loss of the photovoltaic system based on the degree and distribution characteristics of temperature anomalies.

[0036] The mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency is as follows: η = η0× [1 - β × (T - T0) - γ × (Nh / Nt) - δ × (ΔT / T0)] Where η is the actual efficiency, η0 is the efficiency under standard conditions, usually the rated efficiency provided by the manufacturer, β is the temperature coefficient, generally 0.004 / °C, T is the average temperature of the PV panel, T0 is the reference temperature, Nh is the number of hot spots, Nt is the total number of PV panels, ΔT is the difference between the highest temperature and the average temperature, and γ and δ are coefficients related to equipment aging and uneven heat distribution, which are calibrated through experiments.

[0037] This invention can also be expanded to evaluate the efficiency of large-scale ground-based photovoltaic power plants. In this case, the drone platform can adopt a fixed-wing structure, with a flight altitude of 100-150 meters and a single-flight coverage area of ​​50-100 hectares. The system's data processing flow remains unchanged, but the photovoltaic panel recognition algorithm needs to be optimized for the layout of the ground-based photovoltaic arrays, with a particular focus on the effects of terrain variations and inter-array shadowing.

[0038] The mathematical model of the relationship between PV panel temperature distribution and power generation efficiency also needs to be adjusted based on the characteristics of ground-based power stations, adding parameters to account for dust cover and partial shading: η = η0 × [1 - β × (T - T0) - γ × (Nh / Nt) - δ × (ΔT / T0) - ε × D] Where D is the dust cover index, which is calculated by the reflectivity change of visible light images, and ε is the dust influence coefficient, which is calibrated according to the environmental conditions in different regions.

[0039] This embodiment also provides a photovoltaic system efficiency evaluation system based on drone multi-source image fusion, including: The data acquisition subsystem consists of three parts: data acquisition subsystem, data processing subsystem and efficiency evaluation subsystem. The data acquisition subsystem includes a drone platform, a visible light camera, a thermal imaging camera and an environmental parameter acquisition module, which is used to collect visible light orthophotos, thermal imaging data and environmental parameters of the photovoltaic system. The environmental parameter acquisition module includes a temperature sensor, a light intensity sensor and a wind speed sensor. In the data acquisition subsystem, the UAV platform adopts a six-rotor structure, with a load capacity of not less than 3kg, a flight time of not less than 25 minutes, and is equipped with a GPS / RTK positioning module with a positioning accuracy better than 5cm; the visible light camera is a high-resolution camera with not less than 20 million pixels and is equipped with a zoom lens with an equivalent focal length of 24-70mm; the thermal imaging camera has a resolution of not less than 640×480 pixels, a temperature measurement range of -20℃ to 150℃, and a temperature resolution of not higher than 0.05℃; the environmental parameter acquisition module includes a temperature sensor, a light intensity sensor, and a wind speed sensor to record environmental parameters.

[0040] The drone's flight path planning adopts a serpentine scanning path. The flight altitude is determined according to the size and shape of the rooftop photovoltaic system, generally controlled between 30-80 meters, to ensure that the overlap of adjacent routes is not less than 70%, and the overlap of adjacent images on the same route is not less than 80%; during the flight, the visible light camera and the thermal imaging camera shoot synchronously, and the shooting interval is automatically adjusted according to the drone's flight speed to ensure that complete coverage of image data is obtained.

[0041] The data processing subsystem includes a data transmission module, a ground control station, an image preprocessing unit, a multi-source image registration module, and a photovoltaic panel identification and segmentation module. It uses an improved feature point matching algorithm to obtain high-precision spatial registration results. Combining the spatial characteristics of visible light images with the temperature characteristics of thermal imaging, it uses a deep learning model to accurately segment photovoltaic panels and simultaneously identify characteristic information such as panel type, arrangement, and installation angle. The efficiency evaluation subsystem includes a temperature anomaly detection module, an efficiency evaluation module, and a result output module. It is used to establish a mathematical model of the relationship between the temperature distribution of photovoltaic panels and power generation efficiency, intelligently identify and classify temperature anomalies, distinguish between normal operating temperature differences and fault hotspots, and determine the degree of efficiency attenuation and potential fault types of the photovoltaic system through thermal distribution pattern analysis.

[0042] This embodiment can quickly and accurately identify photovoltaic system anomalies and evaluate their operating efficiency, providing a scientific basis for photovoltaic system operation and maintenance management, significantly improving evaluation efficiency and accuracy, and reducing operation and maintenance costs.

[0043] The present invention not only solves the technical difficulties of the existing technology in evaluating the efficiency of rooftop photovoltaic systems, but also provides a scientific basis for the operation and maintenance management of photovoltaic systems. It is of great significance to promote the popularization and application of distributed photovoltaic systems and improve energy utilization efficiency.

[0044] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A photovoltaic system efficiency evaluation method based on multi-source image fusion of UAVs is characterized by: include: S1. Collect visible light orthophotos, thermal imaging data and environmental parameters of the photovoltaic system; S2. Based on the collected visible light orthophotos, thermal imaging data, and environmental parameters, an improved feature point matching algorithm is used to obtain a high-precision spatial registration result between the thermal image and the orthophoto. S3. Based on the registration results and the spatial characteristics of the visible light image and the temperature characteristics of the thermal image, the deep learning model is used to automatically identify the position of the photovoltaic panels, achieve accurate segmentation of the photovoltaic panels, and simultaneously identify the characteristic information of the photovoltaic panel type, arrangement, and installation angle; S4. Establish a mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency, intelligently identify and classify temperature anomalies, distinguish between normal operating temperature differences and fault hotspots, and determine the degree of photovoltaic system efficiency attenuation and potential fault types through thermal distribution pattern analysis.

2. The photovoltaic system efficiency evaluation method based on multi-source image fusion of unmanned aerial vehicles according to claim 1, characterized in that: The step S2 comprises: S21, performing image stitching processing on the collected visible light orthophotos to generate orthophotos; S22. Perform geometric correction and radiation correction on thermal imaging; S23. Combining the edge and corner features of the photovoltaic panel, the ORB algorithm is used to extract feature points, and preliminary matching is performed using the Hamming distance and FLANN fast approximate nearest neighbor method. The RANSAC algorithm is used to eliminate erroneous points and calculate the homography matrix and the basic matrix. Then, based on the improved NCC method, bilinear interpolation and bidirectional verification mechanism are combined to adapt to rotation and scale changes and correct deformations caused by shooting angles and sensor differences. Finally, by dynamically adjusting the search window size, combining the RMSE elimination mechanism and the correlation coefficient threshold to screen matching points, the registration accuracy is better than 2 pixels.

3. The photovoltaic system efficiency evaluation method based on multi-source image fusion of unmanned aerial vehicles according to claim 1, characterized in that: The step S3 comprises: S31. Extract texture and geometric features from visible light orthophotos; S32, combining temperature information from thermal imaging to assist boundary optimization; S33, fusing visible light spatial features with thermal imaging temperature features through a multi-source feature fusion formula; S34. Inputting the fused feature vector into a deep learning model to accurately segment the photovoltaic panels and simultaneously identify feature information including the type, arrangement, and installation angle of the photovoltaic panels; S35. Post-process the recognition results to eliminate noise and misidentification, and generate a photovoltaic panel area mask.

4. The photovoltaic system efficiency evaluation method based on multi-source image fusion of unmanned aerial vehicles according to claim 3 is characterized in that: The multi-source feature fusion formula is as follows: Ffusion = α × FRGB + β × FIR + γ × Fgeo in, Ffusion is the fused feature vector, FRGB is the spatial feature vector extracted from the visible light image, FIR The temperature feature vector extracted from thermal imaging, Fgeo is the geometric shape eigenvector, α 、 β 、 γ is the weight coefficient, and α + β + γ =1.

5. The photovoltaic system efficiency evaluation method based on multi-source image fusion of unmanned aerial vehicles according to claim 3 is characterized in that: The deep learning model loss function is as follows: L total =L seg +λ 1 L cls +λ 2 L reg L total =L seg +λ 1 L cls +λ 2 L reg in, L total is the total loss function, L seg is the segmentation loss, L cls is the classification loss, used for photovoltaic panel type recognition, L reg is the regression loss, used for geometric parameter prediction, λ 1 and λ 2 is the loss weight balance parameter; The Dice Loss segmentation loss is calculated as follows: L seg =1−2∣P∩G∣∣P∣+∣G∣ L seg =1−∣ P ∣+∣ G ∣2∣ P ∩ G ∣ in, P is the set of photovoltaic panel pixels predicted by the model, G is the set of real labeled photovoltaic panel pixels, | P ∩ G ∣ is the number of intersection pixels between the prediction and the true annotation, ∣ P ∣ is the total number of predicted pixels, ∣ G ∣ is the total number of real labeled pixels; The confidence level of photovoltaic panel identification is calculated as follows:

6. Among them, Cpanel is the confidence level of photovoltaic panel identification, Stemp is the temperature feature score, Sshape is the shape feature score, Stexture is the texture feature score, w 1. w 2 and w 3 is the feature weight parameter, b is the bias parameter, e is the base of natural logarithm; The photovoltaic panel installation angle is calculated as follows:

7. Among them, θ The installation tilt angle of the photovoltaic panel, H max is the highest point height of the photovoltaic panel, H min is the lowest point height of the photovoltaic panel, L panel is the length of the photovoltaic panel, and arctan is the inverse tangent function.

8. The photovoltaic system efficiency evaluation method based on multi-source image fusion of unmanned aerial vehicles according to claim 1, characterized in that: The step S4 comprises: S41, establishing a reference temperature area by measuring the normal operating temperature of surrounding non-faulty photovoltaic panels as a benchmark; S42. Analyze the surface temperature distribution of photovoltaic panels through thermal imaging and identify areas with abnormal temperature; S43, using the set temperature anomaly detection threshold to classify the temperature anomaly points into three levels: slight anomaly, moderate anomaly, and severe anomaly; S44. Determine the type of abnormality based on the distribution pattern of the abnormal temperature points; S45. Evaluate the efficiency loss of the photovoltaic system based on the degree and distribution characteristics of temperature anomalies.

9. The photovoltaic system efficiency evaluation method based on UAV multi-source image fusion according to claim 6, characterized in that: The temperature anomaly detection threshold formula is as follows:

10. Among them, T threshold is the temperature anomaly detection threshold, is the average temperature of the reference area, σref is the standard deviation of the reference area temperature, k is the threshold coefficient.

11. The photovoltaic system efficiency evaluation method based on UAV multi-source image fusion according to claim 1, characterized in that: The mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency is as follows: η = η0× [1 - β × (T - T0) - γ × (Nh / Nt) - δ × (ΔT / T0)] Where η is the actual efficiency, η0 is the efficiency under standard conditions, β is the temperature coefficient, T is the average temperature of the photovoltaic panel, T0 is the reference temperature, Nh is the number of hot spots, Nt is the total number of photovoltaic panels, ΔT is the difference between the highest temperature and the average temperature, and γ and δ are coefficients related to equipment aging and uneven heat distribution.

12. The photovoltaic system efficiency evaluation method based on multi-source image fusion of unmanned aerial vehicles according to claim 1, characterized in that: The mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency also includes: When applied to the efficiency evaluation of large-scale ground-based photovoltaic power plants, the mathematical model of the relationship between photovoltaic panel temperature distribution and power generation efficiency is adjusted according to the characteristics of the ground-based power plants, and parameters for dust coverage and local shading factors are added: η = η0 × [1 - β × (T - T0) - γ × (Nh / Nt) - δ × (ΔT / T0) - ε × D] Where D is the dust cover index, which is calculated by the reflectivity change of visible light images, and ε is the dust influence coefficient, which is calibrated according to the environmental conditions in different regions.

13. The photovoltaic system efficiency evaluation method based on UAV multi-source image fusion according to any one of claims 1 to 9, characterized in that: include: It consists of three parts: data acquisition subsystem, data processing subsystem and efficiency evaluation subsystem; The data acquisition subsystem includes a UAV platform, a visible light camera, a thermal imaging camera and an environmental parameter acquisition module, which is used to collect visible light orthophotos, thermal imaging data and environmental parameters of the photovoltaic system; Among them, the environmental parameter acquisition module includes a temperature sensor, a light intensity sensor and a wind speed sensor; The data processing subsystem includes a data transmission module, a ground control station, an image preprocessing unit, a multi-source image registration module, and a photovoltaic panel identification and segmentation module. It uses an improved feature point matching algorithm to obtain high-precision spatial registration results. Combining the spatial characteristics of visible light images with the temperature characteristics of thermal imaging, it uses a deep learning model to accurately segment photovoltaic panels and simultaneously identify characteristic information such as panel type, arrangement, and installation angle. The efficiency evaluation subsystem includes a temperature anomaly detection module, an efficiency evaluation module, and a result output module. It is used to establish a mathematical model of the relationship between the temperature distribution of photovoltaic panels and power generation efficiency, intelligently identify and classify temperature anomalies, distinguish between normal operating temperature differences and fault hotspots, and determine the degree of efficiency attenuation and potential fault types of the photovoltaic system through thermal distribution pattern analysis.

Images