Classified building roof photovoltaic potential estimation method and device based on remote sensing image

Abstract

The invention provides a classified building roof photovoltaic potential estimation method and device based on a remote sensing image, and belongs to the field of artificial intelligence, and the method comprises the steps: constructing a building classification system for photovoltaic potential evaluation, and obtaining building classification information on the remote sensing image, the classification system specifically comprises the steps that buildings are divided into color steel tile roof industrial and commercial buildings, concrete roof industrial and commercial buildings, urban residential buildings and rural residential buildings; building samples corresponding to the four types of buildings are manufactured; based on the building samples, the instance segmentation pre-training model is finely adjusted to extract the total area of four types of building roofs in the to-be-detected area; according to the total roof area of each type of building and the area of the roof to be developed, the total photovoltaic potential of each type of roof and the photovoltaic potential to be developed are evaluated, and the available coefficient is obtained from the roof where the photovoltaic panel is installed.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and in particular relates to a method and device for estimating photovoltaic potential of classified building roofs based on remote sensing images. Background Art

[0002] With the continued development of the distributed photovoltaic industry, the rational planning and utilization of building rooftops has become particularly important for assessing the potential resources for photovoltaic power generation. Currently, there are a large number of building rooftops that are suitable for installing photovoltaic systems but have not yet been utilized. The photovoltaic potential of these rooftops is closely related to the type of building.

[0003] Building data plays a central role in assessing the potential for photovoltaic (PV) rooftops. This data is primarily categorized into three types: total building area for the entire region; building area data divided by geographic grids, which includes not only area information but also spatial distribution; and building footprint data, which is more detailed and includes the specific location, boundaries, and area of ​​individual buildings. The first two types of data are typically derived from official statistics or inferred from other multi-source data. However, obtaining building footprint data on a large scale is more difficult. Therefore, extracting this information using remote sensing imagery is an effective technique, with semantic segmentation and instance segmentation being two key deep learning methods.

[0004] Existing technical methods fall into two main categories. The first category focuses directly on assessing the potential for PV on building rooftops. Problems with these studies include: First, most studies fail to account for the diversity of building types. The suitability and necessity for PV installations vary significantly across different building types. Treating all building rooftops as equally important PV hosts can lead to overestimation of the assessment results. Second, while some studies have considered building function types in their assessments, these often rely on auxiliary data such as points of interest (POIs), areas of interest (AOIs), or urban functional zoning, or use a coarse-grained classification. This auxiliary data can be ambiguous, incomplete, and unreliable. This classification information primarily serves to statistically analyze the potential of different building types, without performing specialized PV resource potential calculations for each building type. Finally, while some studies have designed corresponding utilization coefficients for different building types in specific regions, these coefficients are often based on theoretical analysis, taking into account factors such as local development plans and building distribution, making them unsuitable for generalization to other regions.

[0005] The second category of research focuses on technical methods for building classification. These methods typically build on existing building vector data, leveraging multimodal data to calculate multiple features for individual buildings. They then train machine learning models such as random forests and XGBoost to identify building functions. Problems with these methods include: First, outdated or inaccurate building vector data can reduce the accuracy of building function identification. Second, these classification standards may not be applicable to rooftop PV potential assessment, as PV potential assessment requires consideration of not only the building's functional type but also its morphological characteristics. Summary of the Invention

[0006] To address the above technical issues, the present invention provides a method and device for estimating rooftop photovoltaic potential of classified buildings based on remote sensing images. First, a building classification standard is established based on the needs of rooftop photovoltaic potential assessment. Further, building classification extraction is performed based on remote sensing images, and the utilization coefficients of different types of buildings are obtained. This allows for targeted rooftop photovoltaic potential estimation for different types of buildings, providing more scientific data support for rooftop photovoltaic development planning. The specific technical solution is as follows:

[0007] A method for estimating photovoltaic potential of classified building roofs based on remote sensing images includes the following steps:

[0008] Step 1: Construct a building classification system for photovoltaic potential assessment to obtain building classification information from remote sensing images. The classification system specifically divides buildings into four categories: industrial and commercial buildings with color steel tile roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings.

[0009] Step 2: Prepare building samples corresponding to the four types of buildings;

[0010] Step 3: Based on the building samples, fine-tune the instance segmentation pre-trained model to extract the total roof area of ​​the four types of buildings in the test area;

[0011] Step 4: Based on the total roof area of ​​each type of building, calculate the developed roof area and the roof area to be developed, and evaluate the total photovoltaic potential and the photovoltaic potential to be developed of each type of roof.

[0012] A device for estimating photovoltaic potential of classified building roofs based on remote sensing images, comprising:

[0013] The building classification module constructs a building classification system for photovoltaic potential assessment, which is used to obtain building classification information from remote sensing images. The classification system specifically divides buildings into four categories: industrial and commercial buildings with color steel tile roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings.

[0014] Building extraction module, which produces building samples corresponding to four types of buildings;

[0015] The building roof area calculation module fine-tunes the instance segmentation pre-training model based on the corresponding building samples of the four types of buildings to extract the total roof area of ​​the four types of buildings in the test area;

[0016] The photovoltaic potential calculation module calculates the developed roof area and the undeveloped roof area based on the total roof area of ​​various types of buildings, and evaluates the total photovoltaic potential and undeveloped photovoltaic potential of each type of roof.

[0017] An electronic device comprises: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the described method.

[0018] A computer-readable storage medium stores executable instructions, which, when executed by a processor, enable the processor to implement the method described above.

[0019] The present invention has the following beneficial effects:

[0020] By constructing a building classification system for photovoltaic potential assessment, this paper can more accurately identify and classify different building types, thereby improving the accuracy of photovoltaic potential assessment. This classification system not only considers the functional characteristics of buildings, but also their morphological characteristics, making the assessment results more realistic.

[0021] The method of the present invention reduces the reliance on auxiliary classification data such as POIs, AOIs, and urban functional zoning. By fully utilizing the differences in image features in remote sensing images, such as shape, size, color, texture, and shadow, the building classification information required for photovoltaic potential assessment can be directly obtained from remote sensing images, greatly reducing the complexity of data collection and processing.

[0022] The present invention utilizes the advantages of remote sensing images such as wide coverage, fast updating and high quality, so that photovoltaic potential assessment can be carried out more quickly and widely. This advantage is particularly evident in large-scale assessments.

[0023] This invention provides more scientific data support for rooftop photovoltaic development planning. By accurately evaluating the rooftop photovoltaic potential of different types of buildings, it can provide strong data support for the planning and implementation of photovoltaic projects, thereby improving the success rate and economic benefits of the projects.

[0024] This method uses the area conversion factor derived from rooftops already equipped with photovoltaic panels, avoiding the potential limitations of factors derived from theoretical or empirical calculations. This approach is more realistic and can more accurately reflect the actual photovoltaic installation potential of different building rooftop types. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the overall framework diagram of the method of the present invention;

[0026] Figure 2 Schematic diagram of overlapping sliding windows and ignoring edge models. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other. To achieve the above-mentioned objectives, the present invention adopts the following technical solutions.

[0028] The overall framework of the method for estimating photovoltaic potential of classified building roofs based on remote sensing images proposed in this invention is shown in the figure below: Figure 1 shown.

[0029] Step 1: Construct a building classification system for photovoltaic potential assessment to obtain building classification information from remote sensing images;

[0030] Building classification for photovoltaic potential assessment requires consideration of both functional and morphological characteristics. Based on functional characteristics, rooftop photovoltaics are categorized into two main types: commercial and industrial (C&I) and residential. C&I photovoltaic buildings include industrial plants, commercial buildings, government offices, schools, hospitals, and public buildings. These types of buildings typically have high electricity loads and higher electricity prices, resulting in higher expected returns from photovoltaic projects. Industrial plants, in particular, have large, flat roofs with minimal obstruction and shadowing, making them a preferred choice for photovoltaic development. Residential photovoltaic buildings are primarily used in residential buildings. Due to the complex ownership of rooftops in urban residential buildings, residential photovoltaics are primarily concentrated in rural residential buildings. However, rural residential buildings face challenges such as limited available area, unclear load capacity, and low electricity prices, resulting in relatively limited photovoltaic development potential. Therefore, in terms of the priority of distributed photovoltaic development, C&I buildings have the highest potential for photovoltaic development, followed by rural residential buildings, and finally urban residential buildings.

[0031] Based on building morphology, building roofs are categorized into color-coated steel tile roofs and concrete roofs. Color-coated steel tile roofs are commonly found on low-rise buildings, factories, and warehouses. While these roofs have relatively weak load-bearing capacity, they possess a certain slope, making them suitable for direct installation of photovoltaic modules. Remote sensing image characteristics of these roofs include their shape, size, color, texture, and shading. Concrete roofs are commonly found on high-rise buildings, factories, large commercial buildings, and rural residences. These roofs have a stronger load-bearing capacity and are suitable for mounting photovoltaic modules at optimal angles. Remote sensing image characteristics of these roofs also include shape, size, color, texture, and shading.

[0032] Based on the functional and morphological characteristics of these buildings, we categorize them into four types: industrial and commercial buildings with color-coated steel roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings. This classification system utilizes only differences in remote sensing image features, such as shape, size, color, texture, and shadows, to obtain the building classification information necessary for photovoltaic potential assessment.

[0033] Step 2: Create building samples corresponding to the four types of buildings to extract the four types of buildings;

[0034] This invention provides a method for batch-generating building samples, providing high-quality training data for building classification for photovoltaic potential assessment. This method efficiently generates building samples suitable for the classification system of this invention by combining building footprint data, area of ​​interest (AOI) data, and visual interpretation. The method includes the following steps:

[0035] Data preparation, including: obtaining building footprint data and area of ​​interest (AOI) data;

[0036] Acquiring building footprint data involves: Building footprints are the projected outlines of buildings in a bird's-eye view, providing key information such as their location, boundaries, and floor space. This data is extracted using remote sensing imagery combined with deep learning models, but may contain annotation errors.

[0037] Acquiring Area of ​​Interest (AOI) data involves storing AOI data in polygonal vector form, representing areas with specific functions or significance (such as residential, commercial, or industrial areas), and including relevant attribute information. Using the attributes of the AOI data, you can initially determine the functional type of buildings within the area.

[0038] Sample area selection: This includes overlaying building footprint data with AOI data, visually selecting high-quality areas within the AOI as sample areas, and downloading remote sensing images corresponding to the sample areas for subsequent building footprint correction and sample annotation.

[0039] Building footprint correction: Because building footprint data is extracted from historical remote sensing imagery, it contains two types of errors: distribution noise (labeling errors) and shape noise (imprecise depiction). Therefore, both types of noise need to be corrected. The specific method is: in ArcMap software, the building footprint is moved as a whole relative to the remote sensing image to eliminate the spatial deviation between the building footprint and the remote sensing image; patches mistakenly extracted as buildings are deleted; unlabeled buildings are added; and inaccurate building boundaries are corrected.

[0040] Sample type annotation, including:

[0041] Preliminary functional attribute acquisition: Based on the attribute information of AOI data, the functional type of buildings in the sample area (such as residential, commercial, industrial, public buildings, etc.) is preliminarily determined.

[0042] Visual interpretation correction: Combined with the visual interpretation of remote sensing images, the preliminary functional attributes are further corrected to the four building types defined in this invention, namely, industrial and commercial buildings with color steel tile roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings.

[0043] Based on the four building samples obtained using the above method, the pre-trained instance segmentation model was further fine-tuned to achieve accurate recognition and boundary delineation of different building types. The instance segmentation task requires the algorithm to not only identify the target objects in the image but also accurately distinguish the boundaries of each object. YOLOv8 was selected as the pre-trained model for instance segmentation. Based on the dataset size and device performance, the algorithm was selected from five pre-trained models of different sizes provided by YOLOv8.

[0044] By fine-tuning the selected pre-trained YOLOv8 instance segmentation model using the four building sample sets obtained using the aforementioned method, the model was able to identify different building types, resulting in a fine-tuned instance segmentation model. This model can accurately identify building types and precisely delineate their boundaries, providing technical support for subsequent photovoltaic potential assessment.

[0045] Step 3: The fine-tuned instance segmentation model will be used to extract information of four types of buildings in a large area;

[0046] When applying the trained model to extract information about four types of buildings in a large area, it is necessary to crop the remote sensing image into fixed blocks for extraction and then splice the extracted results. The present invention uses a sliding overlapping window for extraction and draws on the existing strategy of ignoring edges for splicing. Specifically, only the targets that fall within the effective area in the center of the window are retained. For targets that fall exactly on the edge of the effective area, they will only be retained if and only if they fall on the right and bottom edges. Figure 2As shown in the figure, the solid rectangle represents the original remote sensing image, the thin dashed rectangle represents the padded remote sensing image, the thick dashed square (A) represents the sliding window, and the diamond-filled square (a) represents the valid area within the sliding window. The side length of A is set to the size of the Yolov8 instance segmentation model input image, that is, 640, the side length of a is set to 500, the step size of the sliding window is also 500, and the percentage of area a occupying the range of A is set to approximately 0.6. In the final stitching result, each pixel is predicted only once and is not an edge pixel. To ensure that the valid area can completely cover the original image, the original image is padded with 70 pixels on each side before extraction.

[0047] The complete process of large-scale extraction and post-processing includes: (1) Traversing the remote sensing images of each township-level administrative district, using the above-mentioned sliding overlapping window method, using the trained model to extract each window, and recording the information of each window, the bounding box within the window, and the mask. (2) Performing the following processing on each window: reading the pixel coordinates of the sliding window and converting them to longitude and latitude coordinates; calculating the pixel coordinates of the valid area and converting them to longitude and latitude coordinates; using the idea of ​​edge rejection, screening the valid bounding box and converting the pixel coordinates to longitude and latitude coordinates; reading the mask within the valid bounding box. (3) Splicing the results of each window, storing the sliding window, valid area, and bounding box in shpfile format, and the mask in tif format. (4) Using the community AOI and industrial and commercial AOI to perform two-step error correction on the mask. (5) Converting the corrected mask from tif format to shpfile format, setting the projection coordinate system and calculating the area, using the township-level administrative district boundary as the mask, and counting the total area of ​​each type of building roof in each township-level administrative district.

[0048] The aforementioned "two-step error correction" refers to the present invention optimizing the recognition results of the instance segmentation model based on AOI data. Compared to color-coated steel-roofed industrial and commercial buildings and rural residential buildings, some buildings in the concrete-roofed industrial and commercial buildings and urban residential buildings share similar image features. Therefore, AOI data is used to correct confusion between these two categories. The rules are as follows: First, if some buildings fall within the community AOI but are identified as concrete-roofed industrial and commercial buildings, they are corrected to urban residential buildings. Second, if some buildings fall within the industrial and commercial AOI but are identified as urban residential buildings, they are corrected to concrete-roofed industrial and commercial buildings.

[0049] Step 4: Evaluate the untapped photovoltaic potential and total photovoltaic potential of each type of roof;

[0050] After obtaining the total area of ​​the rooftops of different building types in the region, the rooftop area needs to be converted into usable rooftop area and area suitable for installing photovoltaic panels. The conversion from rooftop area to usable rooftop area is done to eliminate shadow areas caused by elevator shafts, parapets, and other factors, as well as the area required for post-installation maintenance of photovoltaic panels. The conversion from usable rooftop area to area suitable for installing photovoltaic panels takes into account the installation method of the photovoltaic panels. Photovoltaic panel installation can be divided into flat installation and optimal tilt installation. The optimal tilt angle refers to the use of brackets to support the photovoltaic panels so that the panels can receive the greatest amount of solar radiation. The optimal tilt angle is determined by latitude. The optimal spacing is determined to maximize the use of roof space while ensuring that the front row of photovoltaic panels does not block the rear row of photovoltaic panels. The optimal spacing is determined by factors such as the installation tilt angle, the roof's slope and gradient, and the size of the photovoltaic panels. If the installation is flat, the area suitable for installing photovoltaic panels is assumed to be equal to the usable rooftop area. In rough estimation, the two are not distinguished. The present invention obtains the utilization coefficient from the rooftop area to the area suitable for installing photovoltaic panels based on practical experience. The specific method is: first, based on the acquired building vectors, the tiles in the remote sensing image are screened to exclude tiles that do not contain buildings. The prepared rooftop PV sample dataset is then used to train and fine-tune HRNet or other pre-trained semantic segmentation models. The trained pre-trained semantic segmentation model is used to identify rooftop PV panels based on tiles. Building roofs can then be classified into two categories: developed roofs and undeveloped roofs, and their areas can be calculated. The area utilization rate of each roof type is calculated by calculating the ratio of the PV panel area to the roof area. For each roof type, the average area utilization rate of all developed roofs in the area is taken as the roof availability factor for that type. Applying the availability factor to the total area of ​​roofs in that type of roof can estimate the total area available for PV installation. Applying the availability factor to the undeveloped roof area within that type of roof can estimate the area available for PV installation. Once the available PV panel area is determined, the installed capacity potential can be calculated using the power generation per unit area of ​​commonly available PV modules. Annual power generation can be estimated in two ways: by multiplying the installed capacity by the annual effective utilization hours in the area; or by multiplying the PV panel area by the solar radiation and system conversion efficiency. The total photovoltaic potential of each type of roof can be calculated based on the total area of ​​photovoltaic panels that can be installed on each type of roof; the photovoltaic potential of each type of roof to be developed can be calculated based on the area of ​​photovoltaic panels that can be installed on each type of roof.

[0051] The present invention provides a device for estimating photovoltaic potential of classified building roofs based on remote sensing images, comprising:

[0052] The building classification module constructs a building classification system for photovoltaic potential assessment, which is used to obtain building classification information from remote sensing images. The classification system specifically divides buildings into four categories: industrial and commercial buildings with color steel tile roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings.

[0053] Building extraction module, which produces building samples corresponding to four types of buildings;

[0054] The building roof area calculation module fine-tunes the instance segmentation pre-training model based on the corresponding building samples of the four types of buildings to extract the total roof area of ​​the four types of buildings in the test area;

[0055] The photovoltaic potential calculation module calculates the developed roof area and the undeveloped roof area based on the total roof area of ​​various types of buildings, and evaluates the total photovoltaic potential and undeveloped photovoltaic potential of each type of roof.

[0056] The present invention also provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the described method.

[0057] The present invention also provides a computer-readable storage medium having executable instructions stored thereon. When the instructions are executed by a processor, the processor is enabled to implement the method described above.

[0058] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk drives, CD-ROMs, optical storage devices, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0059] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0060] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0062] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0063] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for estimating photovoltaic potential of classified building roofs based on remote sensing images, characterized in that: The following steps are involved: Step 1: Construct a building classification system for photovoltaic potential assessment to obtain building classification information from remote sensing images. The classification system specifically divides buildings into four categories: industrial and commercial buildings with color steel tile roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings. Step 2: Prepare building samples corresponding to the four types of buildings; Step 3: Based on the building samples, fine-tune the instance segmentation pre-trained model to extract the total roof area of ​​the four types of buildings in the test area; Step 4: Based on the total roof area of ​​each type of building, calculate the developed roof area and the roof area to be developed, and evaluate the total photovoltaic potential of each type of roof.

2. The method for estimating photovoltaic potential of classified building roofs based on remote sensing images according to claim 1, characterized in that: Step 2 includes: producing four types of building samples based on building footprints, AOI data, and visual interpretation; the building footprint is the bottom outline of the building, which is the outline data of the building projected onto the ground in a bird's-eye view. The geographical location, boundary spatial range, and floor area of ​​a single building are determined by the building footprint; the AOI data is the spatial boundary and related attributes of the regional range stored in the form of polygonal vector data. The regional range includes residential, commercial, industrial, and public buildings.

3. The method for estimating photovoltaic potential of classified building roofs based on remote sensing images according to claim 2, characterized in that: Step 2 includes the following steps: first, the existing building footprint data and AOI data are superimposed, the high-quality area falling within the AOI data is visually selected as the sample area, and the remote sensing image corresponding to the sample area is downloaded. The sample area is corrected to obtain the building sample, and the type information is further added.

4. The method for estimating photovoltaic potential of classified building roofs based on remote sensing images according to claim 1, characterized in that: Step 3 includes: Step 3.1: Fine-tune the instance segmentation pre-trained model based on building samples. Step 3.2: Traverse the remote sensing images of the area to be measured, use sliding overlapping windows, use the trained model to extract each window, and record the information of each window, the bounding box within the effective area of ​​the window, and the mask; Step 3.3, perform the following processing on each window: read the pixel coordinates of the sliding window and convert them to longitude and latitude coordinates; calculate the pixel coordinates of the valid area and convert them to longitude and latitude coordinates; discard the edges, filter the valid bounding box, and convert the pixel coordinates to longitude and latitude coordinates; read the mask within the valid bounding box; Step 3.4: Splice the results of each window, store the sliding window, valid area, and bounding box in shpfile format, and store the mask in tif format; Step 3.5: Use the small area AOI and industrial and commercial AOI to perform two-step error correction on the mask; Step 3.6: Convert the corrected mask from tif format to shpfile format, set the projection coordinate system and calculate the area. Use the township administrative district boundary as the mask and count the total area of ​​the roofs of various types of buildings in the area to be measured.

5. The method for estimating photovoltaic potential of classified building roofs based on remote sensing images according to claim 4, characterized in that: The above two-step error correction means that if a building falls within the community AOI but is identified as a concrete-roofed industrial or commercial building, it will be corrected to an urban residential building; if a building falls within the industrial and commercial AOI but is identified as an urban residential building, it will be corrected to a concrete-roofed industrial or commercial building.

6. The method for estimating photovoltaic potential of classified building roofs based on remote sensing images according to claim 1, characterized in that: In step 4, the roofs of various buildings in the test area are divided into two categories: developed roofs of various types and undeveloped roofs. The area utilization rate of the roof is obtained by calculating the ratio of the photovoltaic panel area on the developed roof to the roof area. For each type of roof, the average area utilization rate of all developed roofs in the area is taken as the utilization coefficient of the roof type. The utilization coefficient is applied to the total area of ​​the roof type to estimate the total area of ​​the roof type that can be installed with photovoltaic panels. The utilization coefficient is applied to the area of ​​roofs to be developed in this type of roof to estimate the area of ​​photovoltaic panels that can be installed on the roofs to be developed; the total photovoltaic potential of each type of roof is calculated based on the total area of ​​photovoltaic panels that can be installed on each type of roof; the photovoltaic potential of each type of roof to be developed is calculated based on the area of ​​photovoltaic panels that can be installed on the roofs to be developed of each type of roof.

7. The method for estimating photovoltaic potential of classified building roofs based on remote sensing images according to claim 6, characterized in that: The roofs of various buildings in the test area are divided into two categories: developed roofs and roofs to be developed. This includes: based on the obtained building roof vectors, the tiles in the remote sensing image are screened to exclude tiles that do not contain buildings. The semantic segmentation model is used to identify rooftop photovoltaic panels based on tiles, and then the building roofs are divided into two categories: developed roofs and roofs to be developed. The areas of each type are calculated, and the area of ​​photovoltaic panels on the developed roofs is also calculated.

8. A device for estimating photovoltaic potential of classified building roofs based on remote sensing images, characterized in that: include: The building classification module constructs a building classification system for photovoltaic potential assessment, which is used to obtain building classification information from remote sensing images. The classification system specifically divides buildings into four categories: industrial and commercial buildings with color steel tile roofs, industrial and commercial buildings with concrete roofs, urban residential buildings, and rural residential buildings. Building extraction module, which produces building samples corresponding to four types of buildings; The building roof area calculation module fine-tunes the instance segmentation pre-training model based on the corresponding building samples of the four types of buildings to extract the total roof area of ​​the four types of buildings in the test area; The photovoltaic potential calculation module calculates the developed roof area and the undeveloped roof area based on the total roof area of ​​various types of buildings, and evaluates the total photovoltaic potential and undeveloped photovoltaic potential of each type of roof.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Executable instructions are stored thereon, and when the instructions are executed by a processor, the processor implements the method according to any one of claims 1 to 7.

Images