A Refined Identification Method and Device for Photovoltaic Power Plants Based on Classifiers and Deep Learning

Abstract

This application belongs to the field of remote sensing detection technology, specifically disclosing a method and apparatus for refined identification of photovoltaic (PV) power plants based on classifiers and deep learning. The method includes: preprocessing multi-source remote sensing images of PV power plants to be identified within a study area; inputting the preprocessed multi-source remote sensing images into a trained machine learning classifier for preliminary coarse-resolution identification, obtaining high-resolution remote sensing images containing coarsely identified outlines of potential PV areas; inputting the high-resolution remote sensing images containing coarsely identified outlines of potential PV areas into a trained deep learning model for refined identification, outputting a segmentation mask with panel type information, wherein the segmentation mask is used to distinguish between PV panels and panel gaps. This application can quickly locate potential areas of PV power plants within a specified area and perform refined identification based on this, improving the accuracy and efficiency of identifying PV power plants in complex and diverse areas.

Description

Technical Field

[0001] This application belongs to the field of remote sensing monitoring technology, and more specifically, relates to a method and device for refined identification of photovoltaic power stations based on classifiers and deep learning. Background Technology

[0002] With increasing emphasis on renewable energy, photovoltaic (PV) power generation, as an important form of clean energy, has experienced rapid growth in both scale and speed of development. As the primary carrier of PV power generation, the operational efficiency and safety of PV power plants directly affect the performance and reliability of the entire PV power generation system. With the continuous increase in the number of PV power plants, the diverse types of equipment, their wide distribution, and complex environments, achieving precise identification and management of PV power plants over a large area is an urgent problem to be solved.

[0003] CN117671513A discloses an optical remote sensing image recognition method, apparatus, device, and storage medium that combines a self-constructed photovoltaic panel index with a recognition model to identify photovoltaic panels. However, this method is prone to losing underlying feature information and cannot distinguish between photovoltaic panels and the gaps between them.

[0004] CN116402109A discloses a method for obtaining a photovoltaic panel recognition model, a photovoltaic panel recognition method, and an apparatus. This method obtains a deep learning model at a specific spatial resolution by creating a training set and its labeled parameters. However, this method requires all remote sensing images of the specified area, resulting in low computational efficiency and making it difficult to meet the needs of rapid recognition of photovoltaic panels across a large area.

[0005] CN112348030A discloses a method, device, remote sensing big data processing platform, and storage medium for identifying solar photovoltaic panels, which outputs pixels containing photovoltaic panels by combining probability thresholds. However, the probability thresholds set by this method still have significant limitations in dealing with complex environments, especially when buildings and greenhouses are confused, and it cannot efficiently identify photovoltaic power stations over a large area. Summary of the Invention

[0006] To address the aforementioned shortcomings of existing technologies, this application provides a method and apparatus for refined identification of photovoltaic power plants based on classifiers and deep learning, aiming to solve the problem of refined identification of photovoltaic power plants over a large area.

[0007] Firstly, this application provides a refined identification method for photovoltaic power plants based on classifiers and deep learning, including:

[0008] Preprocessing of multi-source remote sensing images of photovoltaic power plants to be identified within the study area;

[0009] The preprocessed multi-source remote sensing images are input into a trained machine learning classifier for preliminary coarse-resolution identification, and high-resolution remote sensing images containing coarse outlines of potential photovoltaic areas are obtained.

[0010] High-resolution remote sensing images containing coarse outlines of potential photovoltaic areas are input into a trained deep learning model for fine identification, and a segmentation mask with panel type information is output. The segmentation mask is used to distinguish between photovoltaic panels and panel gaps.

[0011] Secondly, this application also provides a refined identification device for photovoltaic power plants based on classifiers and deep learning, comprising:

[0012] The preprocessing module is used to preprocess the multi-source remote sensing images of the photovoltaic power stations to be identified within the study area;

[0013] The preliminary coarse resolution recognition module is used to input the preprocessed multi-source remote sensing images into a trained machine learning classifier for preliminary coarse resolution recognition, and to obtain high-resolution remote sensing images containing coarse outlines of potential photovoltaic areas.

[0014] The fine-grained recognition module is used to input high-resolution remote sensing images containing coarse outlines of potential photovoltaic areas into a trained deep learning model for fine-grained recognition, and outputs a segmentation mask with panel type information. The segmentation mask is used to distinguish between photovoltaic panels and panel gaps.

[0015] Thirdly, this application also provides a remote sensing big data processing platform, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation thereof.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0017] Fifthly, this application also provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0018] This application provides a method and apparatus for refined identification of photovoltaic (PV) power plants based on classifiers and deep learning. It combines multi-band spectral features of remote sensing image data to perform multi-level identification and refined processing of PV power plants over a large area. By preprocessing image data, optimizing the input parameters of the machine learning classifier, and combining morphological operations and other relevant datasets to filter potential PV areas at coarse resolution, the scope of the area to be predicted is narrowed. A web-based semi-automated refined annotation platform is established using SAM, an automated semantic segmentation tool, and a deep learning model is used to refine the learning of PV panel boundaries and types. A fine-tuned model is established based on land use type, resulting in final identification results with clear boundaries and accurate morphology. This application can quickly locate potential PV power plant areas within a specified region and perform refined identification based on these areas. It minimizes manual intervention in each step, improving the accuracy and efficiency of identifying PV power plants in complex and diverse areas, laying the foundation for obtaining high-quality PV power plant information. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts illustrating the refined identification method for photovoltaic power plants based on classifiers and deep learning provided in this application embodiment;

[0021] Figure 2 This is the second flowchart illustrating the refined identification method for photovoltaic power plants based on classifiers and deep learning provided in this application embodiment;

[0022] Figure 3 This is a schematic diagram of the annotation process of the web-based semi-automatic fine-grained annotation platform provided in this application embodiment;

[0023] Figure 4 This is a schematic diagram of the architecture of the deep learning model provided in the embodiments of this application;

[0024] Figure 5 This is a comparison diagram of the preliminary coarse-resolution recognition result and the refined recognition result provided in the embodiments of this application;

[0025] Figure 6 This is a schematic diagram of the structure of the photovoltaic power station fine identification device based on classifier and deep learning provided in the embodiments of this application;

[0026] Figure 7This is a schematic diagram of the structure of the remote sensing big data processing platform provided in the embodiments of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] Figure 1 This is one of the flowcharts illustrating the refined identification method for photovoltaic power plants based on classifiers and deep learning provided in this application embodiment, such as... Figure 1 As shown, this method is applied to remote sensing big data processing platforms, such as the Google Earth Engine (GEE) platform, and includes at least the following steps:

[0029] S101. Preprocess the multi-source remote sensing images of the photovoltaic power stations to be identified within the study area;

[0030] S102. Input the preprocessed multi-source remote sensing images into the trained machine learning classifier for preliminary coarse-resolution recognition to obtain high-resolution remote sensing images containing coarsely identified outlines of potential photovoltaic areas.

[0031] S103. Input the high-resolution remote sensing image containing the coarse outline of potential photovoltaic areas into the trained deep learning model for fine recognition, and output a segmentation mask with panel type information. The segmentation mask is used to distinguish photovoltaic panels and panel gaps.

[0032] This application utilizes the GEE platform to quickly preprocess multi-source remote sensing images of photovoltaic power plants to be identified within the study area. A machine learning classifier is used to perform preliminary coarse-resolution identification of the photovoltaic power plant areas and background areas within the study area, identifying the coarse outlines of potential photovoltaic regions, narrowing down the area from a large region to a relatively small identification range. Finally, a deep learning model is used to refine the identification of this relatively small range, outputting a segmentation mask with panel type information. This segmentation mask has high spatial resolution and can distinguish between photovoltaic panels and panel gaps, thereby obtaining refined identification results of photovoltaic power plants within the study area.

[0033] Figure 2 This is the second flowchart illustrating the refined identification method for photovoltaic power plants based on classifiers and deep learning provided in this application embodiment. Figure 1 and Figure 2 For S101, the remote sensing big data processing platform can quickly preprocess multi-source remote sensing images of photovoltaic power plants to be identified within the research area.

[0034] The GEE platform is a cloud geospatial computing platform that supports free, petabyte-level remote sensing data, various machine learning algorithms, and shared computing resources, enabling rapid and relatively accurate land cover classification results. Utilizing the GEE platform to process multi-source remote sensing data allows for the integration of the strengths of data from different remote sensing image sources.

[0035] The specific preprocessing process can be as follows: First, cloud removal is performed through the Quality Assurance (QA) band to screen out low-quality remote sensing images; then, the multi-temporal images are merged, and the spatial resolution of the final preprocessed image is the minimum spatial resolution among the multi-source remote sensing images.

[0036] Optionally, the preprocessing process also includes custom image spectral information for subsequent machine learning classification algorithm input, including red, green, and blue visible light bands and near-infrared band information, as well as feature indices such as vegetation index, building index, and custom index.

[0037] For S102, the machine learning classifier on the GEE platform is used to perform preliminary coarse-resolution identification of photovoltaic power plants on the preprocessed multi-source remote sensing images.

[0038] Optionally, the machine learning classifier can be a random forest classifier, a binary classifier, etc., which can classify potential photovoltaic areas and background areas in the input remote sensing image.

[0039] In some embodiments, S102 specifically includes:

[0040] Based on a dataset related to the site selection principles for photovoltaic power plants, the preliminary coarse-resolution identification results output by the machine learning classifier are filtered and eliminated, and morphological operations are performed to obtain remote sensing images containing potential photovoltaic areas.

[0041] The dataset related to the site selection principles for photovoltaic power station construction includes one or more of the following: terrain slope and mountain shadow data, urban nighttime light data, road data, population density data, and solar radiation intensity data.

[0042] Specifically, after the machine learning classifier outputs the classification results of potential photovoltaic areas and background areas in the remote sensing image, it is still necessary to further filter and eliminate them through a dataset related to the photovoltaic construction site selection principle, and optimize them through morphological operations to eliminate noise, so as to obtain the final remote sensing image containing potential photovoltaic areas.

[0043] Optionally, the dataset related to the site selection principles for photovoltaic construction includes one or more of the following: terrain slope and mountain shadow data, urban nighttime light data, road data, population density data, and solar radiation intensity data. Specifically, terrain slope and mountain shadow data are used to calculate terrain slope and mountain shadow; solar radiation intensity data mainly includes data such as solar azimuth and elevation angles; and population density data and urban nighttime light data are used to eliminate identification elements located in urban areas and high-density population areas.

[0044] In some embodiments, the machine learning classifier is trained based on the following steps:

[0045] Acquire remote sensing images containing vector polygon outlines of photovoltaic power stations within the study area;

[0046] Within the vector polygon, PV points are randomly selected uniformly based on the size of the vector polygon;

[0047] Outside the vector polygon, NPV points, which are equal in number to the number of PV points, are randomly selected uniformly based on land use type stratification.

[0048] A machine learning classifier is trained by dividing the training set and validation set based on the selected PV points and NPV points.

[0049] Specifically, using vector polygons of photovoltaic power plants from published literature as the basis for the training dataset, points were randomly selected uniformly from within the vector polygons based on their area, denoted as PV; and points were randomly selected uniformly from outside the vector polygons based on land use type, denoted as NPV, ensuring that the number of PV and NPV points was roughly equal, thus completing the construction of the training dataset. Other parameters remained at the default settings of the GEE platform, and training and validation sets were created to train the machine learning classifier.

[0050] The following example further illustrates the process of preliminary coarse-resolution identification of photovoltaic power stations in this application embodiment.

[0051] Example 1: In this embodiment, multi-source remote sensing imagery based on the GEE platform is used. Vector polygon data of photovoltaic power plants in 2020 from publicly available literature is used as training data. A random forest classifier is selected as the machine learning classifier to obtain preliminary coarse-resolution identification results of photovoltaic power plants in China in 2021. The specific steps include:

[0052] Data preparation phase: Sentinel-2 Harmonized 2A and Landsat 8 SR satellite imagery products were selected as multi-source image sources, and cloud removal was performed using the QA60 band and QA Pixel band, respectively. The bands selected for input to the subsequent machine learning classifier in the two image sets are shown in Table 1.

[0053] Table 1

[0054]

[0055] As shown in Table 1, in the Sentinel-2 Harmonized 2A image set, B2, B3, B4, B8, B11, and B12 represent the blue, green, red, near-infrared, short-infrared 1 (1.614 μm), and short-infrared 2 (2.202 μm) bands, respectively, and B8_stdDev represents the standard deviation of the near-infrared band values ​​over a year. In the Landsat 8 SR image set, SR_B2, SR_B3, SR_B4, SR_B5, SR_B6, and SR_B7 represent the blue, green, red, near-infrared, short-infrared 1 (1.566-1.651 μm), and short-infrared 2 (2.107-2.294 μm) bands, respectively, and EVI represents the enhanced vegetation index, NDBI represents the normalized building index, mNDWI represents the modified normalized differential water index, and NDVI represents the normalized vegetation index.

[0056] The two image sets were merged and the median was calculated to obtain composite images from 2020 and 2021, which also summarize all bands and calculated indices from both image sets, with 21 indices in each composite image. The two composite images from 2020 and 2021 were used for training and prediction of the random forest classifier, respectively.

[0057] Classifier Training Phase: Training data consisted of 30m spatial resolution vector polygon data of 2020 photovoltaic power stations from publicly available literature. Within the vector polygons, a varying number of points were randomly selected based on the polygon area, denoted as PV. Outside the vector polygons, points were randomly selected stratified across different land types according to the ESA WorldCereal 10m land use type product, denoted as NPV. A total of 40,000 PV points and 38,000 NPV points were generated. 75% of the points were randomly allocated to the training set, and 25% to the validation set. The 21 band values ​​of the points in the training set were input into a random forest classifier for training. The number of trees was set to 500, and other parameters were left at the default settings on the GEE platform. The accuracy of the random forest classifier was validated using the validation set, yielding an accuracy of 0.881 and a Kappa coefficient of 0.723.

[0058] Coarse-resolution identification result optimization stage: After obtaining the preliminary coarse-resolution identification results for 2021 using a random forest classifier, further screening is required using datasets related to photovoltaic construction site selection principles. First, the solar azimuth angle is set to 180° and the solar altitude angle to 90°. Slope and mountain shadows are calculated using the Shuttle Radar Topography Mission (30m spatial resolution) terrain dataset, removing identification elements located in areas with slopes greater than 30 degrees and mountain shadows less than 150 mm. Second, since the research object is a large-scale photovoltaic power station, identification elements located in urban and high-density population areas are removed using VIIRS Nighttime urban nighttime light data and WorldPop Global Project Population data. Third, pixels with fewer than 9 adjacent pixels are removed to eliminate noise. Finally, morphological operations are performed on the identification results using a circular kernel focalMax and focalMode filter, with a kernel radius of 1 pixel. Based on the above process, the preliminary coarse-resolution identification results for photovoltaic power stations in 2021 can be obtained and exported as a Shapefile file via the GEE platform. Similarly, preliminary coarse-resolution identification results of photovoltaic power stations for any year can be obtained from the remote sensing image sources used.

[0059] In this embodiment, the preliminary coarse-resolution identification process based on the GEE platform obtains potential photovoltaic areas at coarse resolution by preprocessing image data, optimizing machine learning classifier input parameters, and combining morphological operations and other relevant datasets. This narrows down the area from a large region to the region to be predicted. Therefore, when using deep learning models for prediction, it is not necessary to download all remote sensing images within the study area, thus saving memory space.

[0060] To further improve recognition accuracy, a deep learning model architecture was adopted, specifically optimized for refined photovoltaic power station recognition with high spatial resolution (below 10m).

[0061] For S103, high-resolution remote sensing images containing the outlines of potential photovoltaic areas are input into the trained deep learning model. By learning the boundaries and types of photovoltaic panels in the potential photovoltaic areas, the deep learning model can obtain refined identification results of photovoltaic power stations within the research area.

[0062] In some embodiments, the deep learning model is trained based on the following steps:

[0063] Acquire high-resolution remote sensing images containing the outlines of potential photovoltaic areas;

[0064] Target images were selected from high-resolution remote sensing images. The photovoltaic power stations in the target images have diversity in land use type and panel type, and are evenly distributed within the study area.

[0065] The target image is input into a web-based semi-automatic fine-grained annotation platform based on SAM for prediction, and the segmentation mask is output as the image annotation result.

[0066] A training set is created based on the image annotation results and panel type of the target image;

[0067] Perform data augmentation on the training set to train the deep learning model.

[0068] Specifically, the Segment Anything Model (SAM) enables fast and accurate segmentation of any object in an image or video. During the regular training of deep learning models, sample images are labeled using a web-based semi-automatic fine-grained annotation platform based on SAM.

[0069] Figure 2 This is a schematic diagram of the annotation process of the web-based semi-automatic fine-grained annotation platform provided in this application embodiment, such as... Figure 2 As shown, the platform demonstrates its ability to perform rapid and accurate feature annotation on images of any spatial resolution, with particular advantages for high-resolution imagery, as it can distinguish target features from the background. High-resolution imagery includes drone-captured images and commercial satellite imagery. The platform utilizes the SAM tool, along with user prompts and manual adjustments, to achieve efficient, semi-automated, and detailed annotation.

[0070] In some embodiments, the semi-automated fine-grained annotation process is as follows:

[0071] Step a: Create a high-resolution image to be labeled.

[0072] Users can upload high-resolution images to the platform's code folder. The image format supports TIFF and must contain geospatial information. The image size is customizable, with common options such as 256*256.

[0073] Step b: Prompt for input.

[0074] Users log in and enter the platform's initial interface. The platform loads one image at a time and displays the original image on the left side of the screen. Users can specify areas of interest (foreground) on the original image using prompts such as clicking or selecting boxes, such as photovoltaic panels or buildings. The platform's backend runs the SAM model based on the prompts, performs predictions, generates a segmentation mask, and displays it on the right side of the page. If the foreground that the user wants to annotate does not exist in the displayed image, the user can skip it and proceed directly to the next image for annotation.

[0075] Step c: Model feedback and optimization.

[0076] Users can input foreground and background hints multiple times on the original image. The updated mask changes accordingly and is displayed synchronously on the right side of the page. When the user is satisfied with the annotation results, they can click "Next" to continue annotating the next image. At the same time, the final annotation results are automatically saved as a 0 / 1 TIFF raster image to the platform code folder, with the same name as the original image.

[0077] Optionally, the semi-automated fine-grained annotation platform supports multiple users simultaneously annotating the same batch of images. It sets three statuses in the user annotation status data table: "Annotating," "Completed," and "Pending Annotation." Only images in the "Pending Annotation" status are read, ensuring that multiple users do not annotate the same image repeatedly. The semi-automated fine-grained annotation platform is suitable for annotating fine-grained features such as photovoltaic power plants and buildings, and is well-suited for application scenarios requiring rapid and efficient processing of large volumes of images.

[0078] Furthermore, the specific training process of the deep learning model is as follows:

[0079] First, high-resolution remote sensing images containing the outlines of potential photovoltaic (PV) areas are acquired. Second, a number of target images are selected from these images. The PV sites in these target images exhibit diversity in land use and panel types, and are relatively evenly distributed within the study area. Further, a web-based semi-automatic fine-grained annotation platform, based on the automated image segmentation tool SAM, is pre-built to manually annotate the target images sequentially, creating a training set for the deep learning model. SAM receives user-specified foreground or background points, bounding boxes, or free text prompts, and can generate and update corresponding segmentation masks. Finally, data augmentation operations are performed on the training set to increase the sample size and enhance the model's generalization ability in different scenarios. Simultaneously, the image acquisition time and satellite information are input into the model, and multiple custom technical information outputs can be customized to train the deep learning model.

[0080] In some embodiments, the deep learning model is further fine-tuned based on the following steps:

[0081] Classify the target images in the training set based on land use type, and create training subsets corresponding to different land use types;

[0082] Based on the training subsets corresponding to different land use types, the deep learning models are fine-tuned to obtain the fine-tuned deep learning models corresponding to different land use types.

[0083] Specifically, the deep learning model trained using the aforementioned steps is a general deep learning foundation model. However, different land use types exhibit significant differences in background morphology on satellite images. Therefore, the accuracy of the segmentation mask output by inputting the preliminary coarse-resolution recognition results obtained from the machine learning classifier into the general deep learning foundation model needs to be improved.

[0084] To adapt to the morphological characteristics of photovoltaic power stations under different land use types and improve the recognition accuracy of deep learning models, a general deep learning base model is fine-tuned using land use types to obtain fine-tuned deep learning models corresponding to different land use types. In practical applications, the land use type is pre-determined, and the preliminary coarse-resolution recognition results obtained based on a machine learning classifier are input into the fine-tuned deep learning model for the corresponding land use type.

[0085] The fine-tuning process is as follows: First, the target images in the training set of the aforementioned training process are classified based on land use type to create training subsets corresponding to different land use types; then, using the training subsets corresponding to different land use types, the general deep learning base model is fine-tuned to obtain the fine-tuned deep learning model corresponding to different land use types.

[0086] Optionally, the land use type of the potential photovoltaic areas obtained from the initial coarse-resolution identification process is pre-judged. Then, fine-tuned deep learning models corresponding to different land use types are used to further refine the identification of potential photovoltaic areas for each land use type, ultimately generating an accurate photovoltaic power station distribution map. This distribution map has high spatial resolution, capable of distinguishing between photovoltaic panels and their gaps, laying the foundation for obtaining high-quality photovoltaic power station information and supporting subsequent tasks such as power prediction and emission reduction benefit assessment.

[0087] The following specific example further illustrates the process of refined identification of photovoltaic power stations in this application embodiment.

[0088] Example 2: In this embodiment, the deep learning model adopts an improved DeepLabV3+ architecture.

[0089] In this embodiment, the results obtained in Example 1 are refined for identification, and an attempt is made to simultaneously obtain panel type information. After downloading 256*256 images of the selected area, a total of 6,000 manually labeled samples were obtained through a web-based semi-automatic fine-grained annotation platform, covering 122 photovoltaic power plants. To enable the model to learn the patterns of panel types, panel type information for these 122 power plants was manually collected, including monocrystalline silicon, polycrystalline silicon, and other types.

[0090] Once the annotation labels and panel type information for each image are ready, the deep learning model can be trained. In this embodiment, the DeepLabV3+ architecture and the Xception41 encoder are selected to train the base deep learning model. Figure 3 This is a schematic diagram of the architecture of the deep learning model provided in the embodiments of this application, such as... Figure 3 As shown, the RGB three-channel values ​​of the image are used as model input. After decoding, the one-hot encodings of the image's shooting season and satellite type are sequentially input, and the final output is a segmentation mask with panel type information. During training, the batch size is set to 36, the learning rate is set to 0.001, and the Adam optimizer is used for gradient optimization. The final validation set accuracy is 0.865.

[0091] Because different land use types exhibit significant differences in background morphology on satellite imagery, model fine-tuning and classification training were performed for photovoltaic power stations of different land use types. In this embodiment, fine-tuned deep learning models were established for six land use types (wasteland, farmland, forest, grassland, water bodies, and others). For each fine-tuned deep learning model, the learning rate was set to 0.001, and the Adam optimizer was used to retrain subsamples by traversing the batch size within the interval [32, 48]. The fine-tuned deep learning model with the highest accuracy for each land use type was selected, as shown in Table 2. The land use type dataset comes from the publicly available downloadable 2021 CLCD China Land Cover Type Data, with a spatial resolution of 30m.

[0092] Table 2

[0093]

[0094] As can be seen from Table 2, after model fine-tuning, the accuracy of the validation set in the training subsets for different land use types was improved to varying degrees.

[0095] Figure 4 This is a comparison diagram of the preliminary coarse-resolution recognition result and the refined recognition result provided in the embodiments of this application, as shown in the figure. Figure 4 As shown, the difference between the refined recognition result in this embodiment and the 30m coarse spatial resolution result used in training in 2020 was compared, and it can be seen that the refined recognition result has a more complete recognition effect.

[0096] The photovoltaic (PV) power plant fine-grained identification method based on classifiers and deep learning provided in this application combines multi-band spectral features of remote sensing image data to perform multi-level identification and fine-grained processing of PV power plants over a large area. By preprocessing image data, optimizing the input parameters of the machine learning classifier, and combining morphological operations and other relevant datasets to filter potential PV areas at coarse resolution, the range of areas to be predicted is narrowed. A web-based semi-automated fine-grained annotation platform is established using SAM, an automated semantic segmentation tool, and a deep learning model is used to refine the learning of PV panel boundaries and PV panel types. A fine-tuned model is established based on land use type, resulting in final identification results with clear boundaries and accurate morphology. This application can quickly locate potential PV power plant areas within a specified region and perform fine-grained identification based on this, minimizing manual intervention in each step and improving the accuracy and efficiency of identifying PV power plants in complex and diverse areas, laying the foundation for obtaining high-quality PV power plant information.

[0097] The following describes the photovoltaic power plant fine identification device based on classifiers and deep learning provided in this application. The photovoltaic power plant fine identification device based on classifiers and deep learning described below can be referred to in correspondence with the photovoltaic power plant fine identification method based on classifiers and deep learning described above.

[0098] Figure 6 This is a schematic diagram of the structure of the photovoltaic power station fine-grained identification device based on classifiers and deep learning provided in the embodiments of this application, as shown below. Figure 6 As shown, the device includes at least:

[0099] The preprocessing module 601 is used to preprocess the multi-source remote sensing images of photovoltaic power stations to be identified within the study area.

[0100] The preliminary coarse resolution recognition module 602 is used to input the preprocessed multi-source remote sensing image into the trained machine learning classifier for preliminary coarse resolution recognition, and obtain a high-resolution remote sensing image containing the coarse recognition outline of the potential photovoltaic area.

[0101] The fine-grained recognition module 603 is used to input high-resolution remote sensing images containing coarse outlines of potential photovoltaic areas into a trained deep learning model for fine-grained recognition, and output a segmentation mask with panel type information. The segmentation mask is used to distinguish between photovoltaic panels and panel gaps.

[0102] In some embodiments, the deep learning model is trained based on the following steps:

[0103] Acquire high-resolution remote sensing images containing the outlines of potential photovoltaic areas;

[0104] Target images were selected from high-resolution remote sensing images. The photovoltaic power stations in the target images have diversity in land use type and panel type, and are evenly distributed within the study area.

[0105] The target image is input into a web-based semi-automatic fine-grained annotation platform based on SAM for prediction, and the segmentation mask is output as the image annotation result.

[0106] A training set is created based on the image annotation results and panel type of the target image;

[0107] Perform data augmentation on the training set to train the deep learning model.

[0108] In some embodiments, the deep learning model is further fine-tuned based on the following steps:

[0109] Classify the target images in the training set based on land use type, and create training subsets corresponding to different land use types;

[0110] Based on the training subsets corresponding to different land use types, the deep learning models are fine-tuned to obtain the fine-tuned deep learning models corresponding to different land use types.

[0111] In some embodiments, acquiring remote sensing imagery containing potential photovoltaic areas includes:

[0112] Based on a dataset related to the site selection principles for photovoltaic power plants, the preliminary coarse-resolution identification results output by the machine learning classifier are filtered and eliminated, and morphological operations are performed to obtain remote sensing images containing potential photovoltaic areas.

[0113] The dataset related to the site selection principles for photovoltaic power station construction includes one or more of the following: terrain slope and mountain shadow data, urban nighttime light data, road data, population density data, and solar radiation intensity data.

[0114] In some embodiments, the machine learning classifier is trained based on the following steps:

[0115] Acquire remote sensing images containing vector polygon outlines of photovoltaic power stations within the study area;

[0116] Within the vector polygon, PV points are randomly selected uniformly based on the size of the vector polygon;

[0117] Outside the vector polygon, NPV points, which are equal in number to the number of PV points, are randomly selected uniformly based on land use type stratification.

[0118] A machine learning classifier is trained by dividing the training set and validation set based on the selected PV points and NPV points.

[0119] In some embodiments, the machine learning classifier is a random forest classifier.

[0120] In some embodiments, the deep learning model employs an improved DeepLabV3+ network, which includes a cascaded encoder, decoder, and one-hot encoding module. The encoder is an Xception41 encoder, and the one-hot encoding module is used to perform one-hot encoding on the season in which the input remote sensing image was captured and the type of satellite.

[0121] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.

[0122] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0123] Based on the methods in the above embodiments, this application provides a remote sensing big data processing platform. The device may include: at least one memory for storing programs and at least one processor for executing the programs stored in the memory. When the programs stored in the memory are executed, the processor performs the methods described in the above embodiments.

[0124] Figure 7 This is a schematic diagram of the structure of the remote sensing big data processing platform provided in the embodiments of this application, as shown below. Figure 7 As shown, the remote sensing big data processing platform may include: a processor 701, a communications interface 702, a memory 703, and a communication bus 704. The processor 701, communications interface 702, and memory 703 communicate with each other via the communication bus 704. The processor 701 can call software instructions stored in the memory 703 to execute the methods described in the above embodiments.

[0125] Furthermore, the logical instructions in the aforementioned memory 703 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0126] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0127] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0128] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0129] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0130] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0131] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0132] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A refined identification method for photovoltaic power plants based on classifiers and deep learning, characterized in that, The method, applied to a remote sensing big data processing platform, includes: Preprocessing of multi-source remote sensing images of photovoltaic power plants to be identified within the study area; The preprocessed multi-source remote sensing images are input into a trained machine learning classifier for preliminary coarse-resolution identification, and high-resolution remote sensing images containing coarse outlines of potential photovoltaic areas are obtained. The high-resolution remote sensing image containing the coarse outline of potential photovoltaic areas is input into a trained deep learning model for fine identification, and a segmentation mask with panel type information is output. The segmentation mask is used to distinguish between photovoltaic panels and panel gaps. The deep learning model is trained based on the following steps: Acquire high-resolution remote sensing images containing the outlines of potential photovoltaic areas; Target images are selected from the high-resolution remote sensing images. The photovoltaic power stations in the target images have diverse land use types and panel types, and are evenly distributed within the study area. The target image is input into a web-based semi-automatic fine-grained annotation platform based on SAM for prediction, and the segmentation mask is output as the image annotation result. A training set is created based on the image annotation results and panel type of the target image; Perform data augmentation operations on the training set to train the deep learning model; The machine learning classifier is trained based on the following steps: Acquire remote sensing images containing vector polygon outlines of photovoltaic power stations within the study area; Within the vector polygon, PV points are randomly and uniformly selected based on the size of the vector polygon. Outside the vector polygon, NPV points, which are equal in number to the number of PV points, are randomly and uniformly selected based on land use type stratification. The machine learning classifier is trained by dividing the training set and validation set based on the selected PV points and NPV points. Acquire remote sensing imagery containing potential photovoltaic areas, including: Based on the dataset related to the site selection principles for photovoltaic power station construction, the preliminary coarse-resolution identification results output by the machine learning classifier are filtered and eliminated, and morphological operations are performed to obtain the remote sensing images containing potential photovoltaic areas. The dataset related to the site selection principles for photovoltaic power station construction includes one or more of the following: terrain slope and mountain shadow data, urban nighttime light data, road data, population density data, and solar radiation intensity data.

2. The refined identification method for photovoltaic power plants according to claim 1, characterized in that, The deep learning model is further fine-tuned based on the following steps: The target images in the training set are classified based on land use type, and training subsets corresponding to different land use types are created. Based on the training subsets corresponding to different land use types, the deep learning model is fine-tuned to obtain the fine-tuned deep learning model corresponding to different land use types.

3. The refined identification method for photovoltaic power plants according to claim 1, characterized in that, The machine learning classifier is a random forest classifier.

4. The refined identification method for photovoltaic power stations according to claim 1, characterized in that, The deep learning model employs an improved DeepLabV3+ network, which includes a cascaded encoder, decoder, and one-hot encoding module. The encoder is an Xception41 encoder, and the one-hot encoding module is used to perform one-hot encoding on the shooting season and satellite type of the input remote sensing image.

5. A refined identification device for photovoltaic power plants based on classifiers and deep learning, characterized in that, include: The preprocessing module is used to preprocess the multi-source remote sensing images of the photovoltaic power stations to be identified within the study area; The preliminary coarse resolution recognition module is used to input the preprocessed multi-source remote sensing images into a trained machine learning classifier for preliminary coarse resolution recognition, and to obtain high-resolution remote sensing images containing coarse outlines of potential photovoltaic areas. The fine-grained identification module is used to input the high-resolution remote sensing image containing the coarse outline of potential photovoltaic areas into the trained deep learning model for fine-grained identification and output a segmentation mask with panel type information. The segmentation mask is used to distinguish between photovoltaic panels and panel gaps. The deep learning model is trained based on the following steps: Acquire high-resolution remote sensing images containing the outlines of potential photovoltaic areas; Target images are selected from the high-resolution remote sensing images. The photovoltaic power stations in the target images have diverse land use types and panel types, and are evenly distributed within the study area. The target image is input into a web-based semi-automatic fine-grained annotation platform based on SAM for prediction, and the segmentation mask is output as the image annotation result. A training set is created based on the image annotation results and panel type of the target image; Perform data augmentation operations on the training set to train the deep learning model; The machine learning classifier is trained based on the following steps: Acquire remote sensing images containing vector polygon outlines of photovoltaic power stations within the study area; Within the vector polygon, PV points are randomly and uniformly selected based on the size of the vector polygon. Outside the vector polygon, NPV points, which are equal in number to the number of PV points, are randomly and uniformly selected based on land use type stratification. The machine learning classifier is trained by dividing the training set and validation set based on the selected PV points and NPV points. Acquire remote sensing imagery containing potential photovoltaic areas, including: Based on the dataset related to the site selection principles for photovoltaic power station construction, the preliminary coarse-resolution identification results output by the machine learning classifier are filtered and eliminated, and morphological operations are performed to obtain the remote sensing images containing potential photovoltaic areas. The dataset related to the site selection principles for photovoltaic power station construction includes one or more of the following: terrain slope and mountain shadow data, urban nighttime light data, road data, population density data, and solar radiation intensity data.

6. A remote sensing big data processing platform, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run on the processor, it causes the processor to perform the method as described in any one of claims 1-4.

Images