Roof photovoltaic detection and identification method and system based on satellite remote sensing image

Abstract

The invention relates to the technical field of roof photovoltaic detection, and discloses a roof photovoltaic detection and identification method and system based on a satellite remote sensing image, and the method comprises the steps: carrying out the feature analysis of distributed photovoltaic, and constructing a distributed photovoltaic data set; constructing a YOLOv5 model, and carrying out data preprocessing on the distributed photovoltaic data set; photovoltaic resources are identified through a YOLOv5 model, and an identification result is evaluated and analyzed. The YOLOv5 model is utilized to efficiently process and analyze the satellite remote sensing image, roof photovoltaic resources can be accurately and quickly identified, and the precision and efficiency of photovoltaic detection are improved. Through combination of feature analysis and a deep learning technology, the limitation of traditional manual detection can be effectively overcome, the labor cost is reduced, the detection period is shortened, and meanwhile, the reliability and scientificity of a detection result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of roof photovoltaic detection, and in particular to a roof photovoltaic detection and identification method and system based on satellite remote sensing images. Background Art

[0002] As the global energy crisis intensifies and the demand for sustainable development increases, solar energy, as a clean and renewable energy source, has received widespread attention. In particular, in the promotion and application of rooftop photovoltaic systems, the number of rooftop solar panels installed has increased rapidly, becoming an important part of the urban power system. In order to accurately and efficiently monitor and evaluate the operating status and performance of these distributed photovoltaic systems, it is particularly important to carry out efficient rooftop photovoltaic detection and identification.

[0003] Traditional rooftop photovoltaic detection methods rely on manual inspections or ground measurements, which are inefficient and costly. In remote areas or areas with complex terrain, manual inspections pose great safety risks and difficulties. Therefore, automated detection and identification of photovoltaic resources based on satellite remote sensing images has become a more efficient solution. Remote sensing images can not only provide large-scale, high-precision ground information, but also avoid many inconveniences and safety hazards in manual inspections.

[0004] However, due to the diversity and complexity of rooftop photovoltaic systems, especially under different lighting, climate and environmental conditions, existing remote sensing image analysis methods face many challenges. For example, the shape, color, texture and size characteristics of solar panels vary greatly, and traditional image processing methods are difficult to meet the requirements of high-precision automatic detection.

[0005] In recent years, with the development of deep learning technology, especially the application of convolutional neural network (CNN) in image recognition has made significant progress, target detection algorithms such as YOLO series models have been applied in many fields with good results. As an efficient target detection model, YOLOv5 has become an important tool in the field of image recognition due to its high accuracy, real-time performance and low computing requirements. However, the application of YOLOv5 in photovoltaic resource detection still faces certain challenges, such as data preprocessing, image enhancement, model training and other issues, which need to be further optimized and improved to improve the detection effect in satellite remote sensing images. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is that the existing rooftop photovoltaic resource detection method has the problems of insufficient accuracy, complex data processing, poor environmental adaptability, etc.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for detecting and identifying rooftop photovoltaics based on satellite remote sensing images, comprising:

[0009] Conduct feature analysis on distributed photovoltaics and construct a distributed photovoltaic data set;

[0010] Build the YOLOv5 model and perform data preprocessing on the distributed photovoltaic dataset;

[0011] Photovoltaic resources are identified through the YOLOv5 model, and the identification results are evaluated and analyzed.

[0012] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, wherein: the feature analysis of distributed photovoltaics includes light intensity feature analysis, solar panel shape feature analysis, texture feature analysis, color feature analysis and size feature analysis;

[0013] The illumination intensity feature analysis includes calculating the brightness value of the image and statistically analyzing the brightness distribution of the image;

[0014] The formula for calculating the brightness value of an image is expressed as:

[0015]

[0016] Where L represents the brightness value of the image, R represents the red component of each pixel in the image, G represents the green component of each pixel in the image, and B represents the blue component of each pixel in the image; represents the gradient of the image at position (x, y), and α represents the adjustment parameter;

[0017] The value range of L is [0,255], which indicates the range of brightness;

[0018] The brightness distribution of the statistical image includes calculating the mean value μ of the image brightness value L , standard deviation σ L , maximum value L max and the minimum value L min ;

[0019] The distributed photovoltaic data set includes light intensity features, shape features, texture features, color features and size features.

[0020] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, wherein: the YOLOv5 model includes an input end, a backbone network, a Neck network, and a Head output layer;

[0021] The data preprocessing includes image scaling and padding, image normalization, image data enhancement and label adjustment;

[0022] The image intensity enhancement includes classifying the image according to the illumination intensity feature analysis;

[0023] When μ L ∈[120,180] and σ L ∈[0,30], and L max and L min When the image brightness is not close to the boundary value, the image is in the normal illumination category and photovoltaic panel identification is performed directly;

[0024] When μ L <80 and σ L >40, and L min When it is close to 0, the image is in the backlight category, increasing the brightness and reducing the contrast;

[0025] When L min <80 and σ L >30, and the image brightness is lower than the mean, the image is in the shadow category, and the local brightness and contrast are increased to highlight the details of the photovoltaic panels in the shadow area;

[0026] When L max >220 and σ L When <30, the image is in the bright light category, reducing the local contrast and restoring the details.

[0027] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, wherein: the input end of the YOLOv5 model includes an input image;

[0028] Image preprocessing, Mosaic data enhancement and adaptive anchor box calculation.

[0029] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, the backbone network of the YOLOv5 model is a lightweight convolutional neural network, including CSPDarknet53, Focus structure and fast spatial pyramid pooling module.

[0030] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, wherein: the Neck network of the YOLOv5 model adopts FPN+PAN structure and CSP2_X structure;

[0031] The Head output layer of the YOLOv5 model uses CIOU loss as the loss function of the bounding box, and uses the binary cross entropy loss function to calculate the classification rate and confidence loss.

[0032] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, the photovoltaic resource identification by the YOLOv5 model includes performing image processing and model training on the selected sample data set by the trained YOLOv5 model, and evaluating and verifying the identification result of the model;

[0033] The sample datasets include RSOD dataset, DIOR dataset, Image Net10 image recognition dataset and some Google Earth historical satellite images;

[0034] By training the training set with the YOLOv5 model, two network model weight training files are obtained, the historical best weight best.pt and the most recent weight last.pt, as verification test files.

[0035] As a preferred solution of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in the present invention, wherein:

[0036] Data module, which performs feature analysis on distributed photovoltaics and builds a distributed photovoltaic data set;

[0037] Preprocessing module, builds the YOLOv5 model and performs data preprocessing on the distributed photovoltaic data set;

[0038] The analysis module uses the YOLOv5 model to identify photovoltaic resources and evaluate and analyze the identification results.

[0039] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0040] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any one of the methods of the present invention.

[0041] Beneficial effects of the invention: The rooftop photovoltaic detection and identification method based on satellite remote sensing images provided by the invention can accurately and quickly identify rooftop photovoltaic resources by using the YOLOv5 model to efficiently process and analyze satellite remote sensing images, thereby improving the accuracy and efficiency of photovoltaic detection. By combining feature analysis with deep learning technology, it can effectively overcome the limitations of traditional manual detection, reduce labor costs, shorten the detection cycle, and at the same time improve the reliability and scientificity of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0043] Figure 1 An overall flow chart of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0044] Figure 2 A schematic diagram of a YOLOv5 network architecture of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0045] Figure 3 A schematic diagram of the Focus structure slicing operation process of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0046] Figure 4 A schematic diagram of the Focus structure convolution process of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0047] Figure 5 A Focus structure replacement equivalent diagram of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0048] Figure 6 A schematic diagram of a statistical diagram of annotated information of a method for detecting and identifying rooftop photovoltaics based on satellite remote sensing images provided in the first embodiment of the present invention;

[0049] Figure 7 A schematic diagram of the length, width and position distribution of the predicted labels of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0050] Figure 8 A training set image of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention;

[0051] Fig. 9 A validation set image of a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided by the first embodiment of the present invention;

[0052] Fig.10 A schematic diagram of curves showing how the loss functions of a training set and a validation set vary with the number of iterations for a rooftop photovoltaic detection and identification method based on satellite remote sensing images provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0054] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a rooftop photovoltaic detection and identification method based on satellite remote sensing images, comprising:

[0055] S1: Conduct feature analysis on distributed photovoltaics and construct a distributed photovoltaic dataset.

[0056] Distributed photovoltaic feature analysis; including light intensity feature analysis, solar panel shape feature analysis, texture feature analysis, color feature analysis and size feature analysis, to build a solar panel data set.

[0057] Shape characteristics: generally square or rectangular.

[0058] Textural features: Surfaces may have a smooth texture or a texture that reflects light, which helps improve energy absorption efficiency.

[0059] Color characteristics: Usually black or dark blue, which helps absorb the energy of sunlight.

[0060] Size characteristics: The size and arrangement of solar panels will have a certain impact on the photovoltaic effect of the battery.

[0061] The illumination intensity feature analysis includes calculating the brightness value of the image and statistically analyzing the brightness distribution of the image;

[0062] The formula for calculating the brightness value of an image is expressed as:

[0063]

[0064] Where L represents the brightness value of the image, R represents the red component of each pixel in the image, G represents the green component of each pixel in the image, and B represents the blue component of each pixel in the image; Represents the gradient of the image at position (x, y), and α represents the adjustment parameter.

[0065] The value range of L is [0,255], which represents the range of brightness.

[0066] The brightness distribution of the statistical image includes calculating the mean value μ of the image brightness value L, standard deviation σ L , maximum value L max and the minimum value L min .

[0067] Based on the above characteristics of solar panels, a certain number of solar panel images are screened to build a basic solar panel dataset.

[0068] Furthermore, by comprehensively analyzing the characteristics of distributed photovoltaics, covering multiple dimensions such as shape, texture, color and size, we lay the foundation for building a high-quality solar panel dataset. By screening representative solar panel images, we can ensure the diversity and representativeness of the dataset, thereby providing accurate and comprehensive input data for subsequent model training and photovoltaic resource identification.

[0069] S2: Build the YOLOv5 model and perform data preprocessing on the distributed photovoltaic dataset.

[0070] The YOLOv5 model is a network based on single-stage target detection. Its feature extraction principle is mainly based on convolutional neural network (CNN), including input end, backbone network, neck network, head output layer. Its detailed structure is as follows: Figure 2 As shown; data preprocessing includes image scaling and padding, image normalization, image data enhancement and label adjustment.

[0071] The image intensity enhancement includes classifying the image according to the illumination intensity feature analysis;

[0072] When μ L ∈[120,180] and σ L ∈[0,30], and L max and L min When the image brightness is not close to the boundary value, the image is in the normal illumination category and photovoltaic panel identification is performed directly;

[0073] When μ L <80 and σ L >40, and L min When it is close to 0, the image is in the backlight category, increasing the brightness and reducing the contrast;

[0074] When L min <80 and σ L >30, and the image brightness is lower than the mean, the image is in the shadow category, and the local brightness and contrast are increased to highlight the details of the photovoltaic panels in the shadow area;

[0075] When L max >220 and σ L When <30, the image is in the bright light category, reducing the local contrast and restoring the details.

[0076] The input of the YOLOv5 model represents the input image, which includes image preprocessing (i.e. data preprocessing), shrinking or filling the image to a specific size, and performing a series of operations such as classification index, center coordinates, width and height normalization.

[0077] The input includes Mosaic data enhancement and adaptive anchor box calculation.

[0078] Mosaic data enhancement is an improvement based on the CutMix data enhancement method. CutMix can only use two pictures for stitching, while Mosaic can randomly scale, crop and arrange four pictures, and freely stitch them together to randomly combine several pictures into one. It can not only enrich the data set, but also enhance the robustness of the model, help improve the performance of small target detection, reduce memory requirements, and greatly increase the training speed.

[0079] The adaptive anchor box calculation is to input image data of a specified size into the network for training, which usually requires scaling or padding the image. YOLOv5 will first calculate the scaling ratio, then calculate the scaled image size, and distribute the padding area evenly to both sides through subtraction and modulus operations, thereby increasing the inference speed.

[0080] The backbone network of the YOLOv5 model is a lightweight convolutional neural network, including CSPDarknet53, Focus structure and fast spatial pyramid pooling module (SPPF); the above network consists of multiple convolutional layers and pooling layers to extract the features of the input image. The design of the backbone network aims to balance the network complexity and feature extraction capabilities.

[0081] CSPDarknet53 is a deep convolutional neural network based on the Darknet architecture with strong feature extraction capabilities; it consists of a series of convolutional layers, pooling layers, and activation functions, which are used to extract and learn the semantic features of images layer by layer; CSPDarknet53 contains multiple residual blocks, which are composed of convolutional layers and residual connections.

[0082] Through the CSPDarknet53 backbone network, YOLOv5 can extract multi-scale and multi-level feature representations from remote sensing images. The above features can be used for subsequent recognition work to achieve the purpose of identifying power facilities or other targets in remote sensing images.

[0083] The Focus structure uses a slice operation to sample the input image to obtain four complementary images; then the feature image channels are overlapped (concat) to expand the number of channels, from the original RGB three channels to 12 channels; then it is convolved to obtain a double-downsampled feature map without losing the original feature image information.

[0084] The Focus structure slicing operation process is as follows Figure 3 As shown, the Focus structure convolution process is as follows Figure 4 As shown, the Focus structure replaces the equivalent structure as Figure 5 shown.

[0085] The Focus structure requires specific devices to run. In the YOLOv5 algorithm, it is replaced with a 6*6 convolutional layer, which can run efficiently on some GPU devices.

[0086] YOLOv5 designs two CSP structures, namely CSP1_X and CSP2_X. CSP1_X is used in Backbone and CSP2_X is used in Neck.

[0087] The Neck network of the YOLOv5 model adopts the FPN+PAN structure and the CSP2_X structure. FPN conveys strong semantic features from top to bottom, while PANet (path aggregation network) conveys positioning features from bottom to top, thereby obtaining a predicted feature map and enhancing the network feature fusion capability.

[0088] The Head output layer of the YOLOv5 model uses CIOU loss as the loss function of the bounding box, and uses the binary cross entropy loss function to calculate the classification rate and confidence loss.

[0089] The YOLOv5 model uses a fully convolutional network structure, which mainly includes multiple convolutional layers, pooling layers, and normalization layers. These layers play a key role in feature extraction. The YOLOv5 model also uses multi-scale prediction, that is, target detection is performed on the input image at different levels of the network, which helps the model to effectively detect targets at different scales. The YOLOv5 model's network structure also uses residual connections and attention mechanisms, which play a certain role in improving the prediction effect.

[0090] After inputting the basic solar panel dataset, the YOLOv5 model goes through a series of preprocessing to improve the recognition effect and accuracy.

[0091] The main steps of image preprocessing include:

[0092] Scaling and padding: YOLOv5 requires the input image size to be fixed, so the image will be scaled or padded to the specified input size, usually square; the processed input size is 640*640 pixels; the scaling process will maintain the aspect ratio of the original image, while the padding process will add pixels (usually black) to the edges of the image to achieve the required size.

[0093] Normalization: The pixel values ​​of an image are usually in the range of 0 to 255, and when training a neural network, it is usually necessary to normalize the input data to a smaller range (such as 0 to 1 or -1 to 1); this helps improve the training stability and performance of the model.

[0094] Data augmentation: It is a widely used preprocessing technique that transforms the original image to generate more usable images and can increase the generalization ability of the model.

[0095] Label adjustment: Along with image preprocessing, YOLOv5 also adjusts the image labels (i.e., the bounding boxes and categories of the objects) to match the results of image preprocessing; for example, if the image is scaled or padded, then the coordinates of the bounding box also need to be adjusted accordingly; if the image is flipped, then the position of the bounding box also needs to be flipped.

[0096] Furthermore, scaling and padding operations are used to ensure that the input images have a uniform size (usually 640x640 pixels) while maintaining the aspect ratio of the image to avoid information loss. The normalization operation scales the pixel values ​​of the image to a range of 0 to 1 or -1 to 1, which helps to improve the stability and convergence speed of network training. Data augmentation effectively increases the diversity of training samples by performing random transformations on images (such as rotation, cropping, flipping, etc.), enhances the generalization and robustness of the model, and performs well in processing small targets. Label adjustment is performed simultaneously with image preprocessing to ensure that the bounding box and target category of the image can match the changes in the image after preprocessing, thereby ensuring accuracy and consistency during training. Through these refined preprocessing steps, the YOLOv5 model can better detect targets on distributed photovoltaic datasets and improve recognition accuracy and efficiency.

[0097] S3: Identify photovoltaic resources through the YOLOv5 model and evaluate and analyze the identification results.

[0098] The selected sample datasets include the RSOD dataset, the DIOR dataset, the Image Net10 image recognition dataset, and some Google Earth historical satellite images.

[0099] The highest resolution of the images in the above dataset can reach 0.3 meters. After image segmentation, data processing, data annotation, format conversion and other operations, 1833 images of 800*800 pixels are obtained. Due to the limitation of the number of datasets, 1528 images are randomly divided as training sets, and 305 images are input into the YOLOv5 model as validation sets. No test set is divided separately. The performance of the validation set is used as the performance prediction of the final model.

[0100] The learning rate set during model training is 0.01, the momentum coefficient is 0.937, the weight decay coefficient is 0.0005, the target box loss weight is 0.05, the classification loss weight is 0.5, the confidence loss weight is 1.0, the IOU threshold is 0.2, the anchor box threshold is 4.0, the batch size is 8, the number of training iterations is 100, and the image size is 640*640 pixels.

[0101] After image preprocessing, the output visual image annotation information is as follows Figure 6 and Figure 7 As shown in the figure, the input image data is made more standardized and more significant, which is beneficial to the subsequent model training.

[0102] Model training: Through the training of the training set by the YOLOv5 model, two network model weight training files are obtained, the best historical weight best.pt and the most recent weight last.pt, which are used for subsequent verification and testing; Figure 8 Shown are images from the training set.

[0103] Based on the YOLOv5 deep learning framework, a target segmentation model for solar panels was trained through 5284 pictures. Its verification set detection accuracy was 0.974. It can efficiently and accurately identify and segment the solar panel area, and then analyze the segmented area to accurately calculate the area occupied by the solar panel and its length, width and other information.

[0104] Calculation results and evaluation: The model predicts that Fig. 9 As shown in the validation set image, the recognition effect of solar panels is good. All targets are fully detected and have appropriate bounding boxes and correctly classified name labels.

[0105] In the YOLOv5 model deep learning, the loss function decline curve is used to observe the model training situation; Fig.10The following are curves showing the changes of the loss functions of the training set and the validation set with the number of iterations. In this experiment, warm-up training was first performed to make the model training more stable in the initial stage. From the beginning of training, there were lower bounding box loss, segmentation loss, confidence loss, and classification loss. As the number of iterations increased, the values ​​of each loss function continued to decrease, and finally stabilized and remained at a lower loss function value, indicating that the model has been effectively learned.

[0106] In this model training, the model's prediction effect is the best at the 98th iteration. Table 1 is a summary of the loss function values ​​when the prediction results are the best.

[0107] Table 1 Summary of loss function values

[0108] Bounding Box Loss Segmentation loss Confidence loss Classification Loss Training set 0.01516 0.01092 0.01470 0.00023 Validation set 0.01375 0.01727 0.01559 6e-05

[0109] Furthermore, by using the YOLOv5 model for automatic identification and evaluation of photovoltaic resources, the solar panel area can be efficiently and accurately identified and segmented, thus providing data support for subsequent resource evaluation and planning. By integrating and processing multiple data sets and combining deep learning technology, the model can learn the characteristics of solar panels from a large number of images and achieve high-precision recognition on the test set (the verification set accuracy reaches 0.974). The loss function decline curve during the model training process shows that the model is gradually optimized during the training process, the learning effect is good, and it can eventually accurately classify and locate the target stably.

[0110] Embodiment 2 is an embodiment of the present invention, which provides a rooftop photovoltaic detection and identification system based on satellite remote sensing images, including:

[0111] The data module performs feature analysis on distributed photovoltaics and constructs a distributed photovoltaic data set.

[0112] The preprocessing module builds the YOLOv5 model and performs data preprocessing on the distributed photovoltaic dataset.

[0113] The analysis module uses the YOLOv5 model to identify photovoltaic resources and evaluate and analyze the identification results.

[0114] Embodiment 3, an embodiment of the present invention, is different from the first two embodiments in that:

[0115] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0116] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0117] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0118] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0119] Example 4 is an embodiment of the present invention, which provides a rooftop photovoltaic detection and identification method and system based on satellite remote sensing images. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0120] A region rich in distributed photovoltaic resources was selected as the test area, and image data covering the region was obtained through satellite remote sensing images with a resolution of 0.3 meters. The sample dataset consists of the RSOD dataset, the DIOR dataset, the ImageNet10 image recognition dataset, and some Google Earth historical satellite images. 1,833 800×800 pixel images were selected, of which 1,528 were used as training sets and 305 were used as validation sets. The test set was not divided separately, and the performance of the validation set was used as the evaluation criterion for the final performance of the model. The shape, texture, color, and size characteristics of solar panels in remote sensing images were analyzed:

[0121] Shape characteristics: Analysis found that the vast majority of solar panels are rectangular or square, with a small number being tilted.

[0122] Texture features: The surface of the solar panel has a characteristic texture of regularly arranged lattices or smooth reflections.

[0123] Color characteristics: Dark blue or black are the main colors of solar panels.

[0124] Size characteristics: Standard panels range in length from 1.6 to 2.2 meters and in width from 1 to 1.2 meters.

[0125] Based on the above features, remote sensing image segmentation technology is used to annotate the dataset and generate training samples containing target box coordinates and classification labels.

[0126] The YOLOv5 model is used, and its structure includes input end, backbone network, Neck network and Head output layer.

[0127] Input: Scale and pad the dataset to unify the image size to 640×640 pixels; use Mosaic data enhancement to randomly crop and splice images to improve the model’s robustness to complex scenes; and calculate the adaptive anchor frame to optimize the matching effect of the target frame.

[0128] Backbone network: Use CSPDarknet53 to extract multi-level features, combine the Focus structure to improve the expression ability of input information, and integrate multi-scale information through the fast spatial pyramid pooling module (SPPF).

[0129] Neck network: It adopts FPN+PAN structure to realize the upper and lower fusion of semantic features and positioning features, and introduces CSP2_X structure to enhance feature expression ability.

[0130] Head output layer: Use CIOU loss to calculate the bounding box loss, and combine the binary cross entropy loss function to optimize the classification rate and confidence.

[0131] The learning rate is set to 0.01, the momentum coefficient is set to 0.937, the weight decay coefficient is set to 0.0005, the batch size is set to 8, and the model training is completed after 100 iterations. The best historical weight file best.pt and the most recent weight file last.pt are saved for subsequent verification. The trained model is used for verification set detection, and the measured accuracy is 97.4%. The experimental results are shown in Table 2.

[0132] Table 2 Experimental results

[0133]

[0134]

[0135] As can be seen from the table, the YOLOv5 model used in the present invention is significantly superior to traditional image processing algorithms and other target detection algorithms in terms of accuracy, recall rate, F1 value, detection speed and false detection rate. Compared with traditional image processing algorithms, the accuracy of the YOLOv5 model has increased by 12.7 percentage points, and the false detection rate has decreased by 6.3 percentage points, which shows that the model has a stronger ability to identify photovoltaic resources and can locate targets more accurately under complex backgrounds. Compared with Faster R-CNN and YOLOv4, the YOLOv5 model of the present invention is not only more accurate (increased by 7.1 and 3.9 percentage points respectively), but also has a significant advantage in detection speed (increased by 95ms and 10ms respectively). This is due to the lightweight design and efficient feature fusion structure of YOLOv5, such as the Focus structure and CSP module, which make it still have high robustness in complex scenes.

[0136] In addition, the low false positive rate of the YOLOv5 model (only 2.3%) further demonstrates the optimization effect of the data preprocessing method, including Mosaic data enhancement and adaptive anchor box calculation. The F1 value reaches 96.0%, indicating that the model has achieved a good balance between accuracy and completeness.

[0137] Compared with the existing technology, the present invention significantly improves the automation and accuracy of photovoltaic resource detection in remote sensing images by combining distributed photovoltaic feature analysis with deep learning models, while shortening the detection time, reflecting great innovation and technical advantages. This method not only improves the efficiency of photovoltaic detection, but also provides a scientific basis for power planning and management.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting and identifying rooftop photovoltaics based on satellite remote sensing images, characterized in that: include: Conduct feature analysis on distributed photovoltaics and construct a distributed photovoltaic data set; Build the YOLOv5 model and perform data preprocessing on the distributed photovoltaic dataset; Photovoltaic resources are identified through the YOLOv5 model, and the identification results are evaluated and analyzed.

2. The method for detecting and identifying rooftop photovoltaic power generation based on satellite remote sensing images according to claim 1, characterized in that: The characteristic analysis of distributed photovoltaics includes light intensity characteristic analysis, solar panel shape characteristic analysis, texture characteristic analysis, color characteristic analysis and size characteristic analysis; The illumination intensity feature analysis includes calculating the brightness value of the image and statistically analyzing the brightness distribution of the image; The formula for calculating the brightness value of an image is expressed as: Where L represents the brightness value of the image, R represents the red component of each pixel in the image, G represents the green component of each pixel in the image, and B represents the blue component of each pixel in the image; represents the gradient of the image at position (x, y), and α represents the adjustment parameter; The value range of L is [0,255], which indicates the range of brightness; The brightness distribution of the statistical image includes calculating the mean value μ of the image brightness value L , standard deviation σ L , maximum value L max and the minimum value L min ; The distributed photovoltaic data set includes light intensity features, shape features, texture features, color features and size features.

3. The method for detecting and identifying rooftop photovoltaic power generation based on satellite remote sensing images as claimed in claim 2, characterized in that: The YOLOv5 model includes an input end, a backbone network, a Neck network, and a Head output layer; The data preprocessing includes image scaling and padding, image normalization, image data enhancement and label adjustment; The image intensity enhancement includes classifying the image according to the illumination intensity feature analysis; When μ L ∈[120,180] and σ L ∈[0,30], and L max and L min When the image brightness is not close to the boundary value, the image is in the normal illumination category and photovoltaic panel identification is performed directly; When μ L <80 and σ L >40, and L min When it is close to 0, the image is in the backlight category, increasing the brightness and reducing the contrast; When L min <80 and σ L >30, and the image brightness is lower than the mean, the image is in the shadow category, and the local brightness and contrast are increased to highlight the details of the photovoltaic panels in the shadow area; When L max >220 and σ L When <30, the image is in the bright light category, reducing the local contrast and restoring the details.

4. The method for detecting and identifying rooftop photovoltaic power generation based on satellite remote sensing images as claimed in claim 3, characterized in that: The input end of the YOLOv5 model includes an input image; Image preprocessing, Mosaic data enhancement and adaptive anchor box calculation.

5. The method for detecting and identifying rooftop photovoltaic power generation based on satellite remote sensing images as claimed in claim 4, characterized in that: The backbone network of the YOLOv5 model is a lightweight convolutional neural network, including CSPDarknet53, Focus structure and fast spatial pyramid pooling module.

6. The method for detecting and identifying rooftop photovoltaic power generation based on satellite remote sensing images as claimed in claim 5, characterized in that: The Neck network of the YOLOv5 model adopts FPN+PAN structure and CSP2_X structure; The Head output layer of the YOLOv5 model uses CIOU loss as the loss function of the bounding box, and uses the binary cross entropy loss function to calculate the classification rate and confidence loss.

7. The method for detecting and identifying rooftop photovoltaic power generation based on satellite remote sensing images as claimed in claim 6, characterized in that: The identification of photovoltaic resources through the YOLOv5 model includes image processing and model training of the selected sample data set through the trained YOLOv5 model, and evaluation and verification of the identification results of the model; The sample datasets include RSOD dataset, DIOR dataset, Image Net10 image recognition dataset and some Google Earth historical satellite images; By training the training set with the YOLOv5 model, two network model weight training files are obtained, the historical best weight best.pt and the most recent weight last.pt, as verification test files.

8. A rooftop photovoltaic detection and identification system based on satellite remote sensing images using the method according to any one of claims 1 to 7, characterized in that: Data module, which performs feature analysis on distributed photovoltaics and builds a distributed photovoltaic data set; Preprocessing module, builds the YOLOv5 model and performs data preprocessing on the distributed photovoltaic data set; The analysis module uses the YOLOv5 model to identify photovoltaic resources and evaluate and analyze the identification results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the rooftop photovoltaic detection and identification method based on satellite remote sensing images described in any one of claims 1 to 7 are implemented.

Images