A panoramic map pull-up detection method, device and equipment and a storage medium

Abstract

This invention discloses a method, apparatus, device, and storage medium for detecting mosaic effects in panoramic maps. The method includes: acquiring a panoramic map of a photovoltaic power station and a digital surface model corresponding to the panoramic map; segmenting at least one component region image to be processed from the panoramic map based on the digital surface model; extracting features from each component region image to be processed, and determining an initial mosaic region image based on the extracted features; inputting the initial mosaic region image into a pre-constructed component segmentation neural network model, and determining the target mosaic region in the initial mosaic region image based on the output target segmentation image. The technical solution of this invention solves the problem of accurately locating areas with mosaic effects during the stitching process of existing panoramic electronic maps, improving the accuracy and precision of mosaic region location in panoramic maps, and enabling maintenance personnel to promptly re-acquire and mosaic repair images of mosaic regions.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, apparatus, device, and storage medium for detecting panoramic map decals. Background Technology

[0002] With increasing national investment in the development and research of clean and renewable energy, the photovoltaic industry based on solar energy is developing rapidly. The continuous expansion of photovoltaic power plants has made power plant operation and maintenance an indispensable part of power plant operation.

[0003] Currently, drones are commonly used for the operation and maintenance of photovoltaic power plants. For example, drones can carry cameras to collect images of photovoltaic modules in a power plant, and image analysis-based detection methods can be used to detect and locate module faults. Before inspecting the entire power plant with a drone to detect module faults, the collected images are usually stitched together to generate a panoramic electronic map of the photovoltaic power plant. This facilitates the subsequent location of detected faulty modules on the panoramic electronic map, allowing maintenance personnel to find the corresponding faulty modules for maintenance.

[0004] In the process of stitching panoramic electronic maps, the registration of adjacent images is usually achieved directly by image registration algorithms. However, due to the high similarity of photovoltaic module images, the image registration work is quite difficult. This can also cause some module areas in the stitched panoramic electronic map to have streaks, which in turn affects the stitching quality of the panoramic electronic map and reduces the accuracy of module fault detection and positioning.

[0005] Therefore, accurately detecting areas with artifacts in panoramic electronic maps so that maintenance personnel can promptly re-acquire and mosaic the artifacted areas, thereby improving the stitching quality of panoramic electronic maps and ensuring the accuracy of component fault detection and location, is an urgent problem to be solved in this field. Summary of the Invention

[0006] This invention provides a panoramic map pattern detection method, device, equipment, and storage medium. It combines image feature extraction with a neural network model to accurately detect and locate areas with pattern phenomena in panoramic maps, improving the accuracy and precision of locating pattern areas in panoramic maps. This allows maintenance personnel to promptly re-acquire and mosaic the pattern areas.

[0007] In a first aspect, embodiments of the present invention provide a panoramic map decal detection method, comprising:

[0008] Obtain a panoramic map of the photovoltaic power station, as well as a digital surface model corresponding to the panoramic map;

[0009] Based on the digital terrain model, at least one component area image to be processed is determined by segmentation from the panoramic map.

[0010] Feature extraction is performed on the images of each component region to be processed, and the initial latte art region image is determined based on the extracted features;

[0011] The initial latte art region image is input into a pre-built component segmentation neural network model, and the target latte art region in the initial latte art region image is determined based on the output target segmentation image.

[0012] Secondly, embodiments of the present invention also provide a panoramic map decal detection device, comprising:

[0013] The map model acquisition module is used to acquire a panoramic map of the photovoltaic power station and a digital surface model corresponding to the panoramic map.

[0014] The image to be processed determination module is used to determine at least one component area image from the panoramic map based on the digital terrain model.

[0015] The initial image determination module is used to extract features from the images of each component region to be processed, and determine the initial latte art region image based on the extracted features.

[0016] The latte art region determination module is used to input the initial latte art region image into a pre-built component segmentation neural network model, and determine the target latte art region in the initial latte art region image based on the output target segmentation image.

[0017] Thirdly, embodiments of the present invention also provide a panoramic map decal detection device, the panoramic map decal detection device comprising:

[0018] At least one processor; and

[0019] A memory that is communicatively connected to at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor can implement the panoramic map decal detection method of any embodiment of the present invention.

[0021] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions, which are used to cause a processor to execute the panoramic map latte art detection method of any embodiment of the present invention.

[0022] This invention provides a panoramic map pattern detection method, apparatus, device, and storage medium. The method involves acquiring a panoramic map of a photovoltaic power station and a corresponding digital land surface model. Based on the digital land surface model, at least one component region image to be processed is segmented from the panoramic map. Features are extracted from each component region image to determine an initial pattern region image. The initial pattern region image is then input into a pre-constructed component segmentation neural network model, and the target pattern region in the initial pattern region image is determined based on the output target segmentation image. By adopting the above technical solution, after obtaining the panoramic map and the corresponding digital surface model, information is first extracted and segmented from the panoramic map based on the digital surface model. The areas containing photovoltaic modules to be processed are extracted. Then, by extracting image features from each area to be processed, the initial images of areas where there may be "streaking" are identified. Combined with a pre-trained component segmentation neural network model, the initial images of the streaking areas are segmented, allowing for more refined segmentation. Based on the segmented target images, the target streaking areas with streaking are accurately determined. This solves the problem of accurately locating areas with streaking in existing panoramic electronic map stitching processes, improving the accuracy and precision of streaking area location in panoramic maps. This allows maintenance personnel to promptly re-acquire and mosaic the streaking areas, thereby improving the accuracy of component fault detection based on panoramic maps.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a panoramic map decal detection method according to Embodiment 1 of the present invention;

[0026] Figure 2 This is a flowchart of a panoramic map decal detection method according to Embodiment 2 of the present invention;

[0027] Figure 3 This is an example diagram of a first region set in Embodiment 2 of the present invention;

[0028] Figure 4 This is an example diagram of a component area to be processed in Embodiment 2 of the present invention;

[0029] Figure 5 This is an example diagram of a long straight line subset in Embodiment 2 of the present invention;

[0030] Figure 6 This is an example diagram of a short straight line subset in Embodiment 2 of the present invention;

[0031] Figure 7 This is a flowchart illustrating a training method for a component segmentation neural network model according to Embodiment 2 of the present invention.

[0032] Figure 8 This is an example diagram of the network structure of a component segmentation neural network model in Embodiment 2 of the present invention.

[0033] Figure 9 This is an example image of a target segmentation image according to Embodiment 2 of the present invention;

[0034] Figure 10 This is an example diagram of a target latte art area positioning frame according to Embodiment 2 of the present invention;

[0035] Figure 11 This is a schematic diagram of the structure of a panoramic map decal detection device according to Embodiment 3 of the present invention;

[0036] Figure 12 This is a schematic diagram of the structure of a panoramic map decal detection device according to Embodiment 4 of the present invention. Detailed Implementation

[0037] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0038] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0039] Example 1

[0040] Figure 1 This is a flowchart of a panoramic map pattern detection method provided in Embodiment 1 of the present invention. The present invention can be applied to the detection, location and identification of areas with pattern phenomena in a panoramic map stitched from multiple images. The method can be executed by a panoramic map pattern detection device, which can be configured on a computer device, such as a laptop, desktop computer or smart tablet.

[0041] like Figure 1 As shown in the figure, the panoramic map decal detection method provided in this embodiment includes the following steps:

[0042] S101. Obtain a panoramic map of the photovoltaic power station and a digital land model corresponding to the panoramic map.

[0043] In this embodiment, a panoramic map can be specifically understood as an electronic map obtained by stitching together images of different sub-scenes within the same large scene after taking overlapping multi-angle photographs using an image capturing device. A Digital Surface Model (DSM) can be specifically understood as a ground elevation model that includes the heights of surface buildings, bridges, and other features.

[0044] Specifically, when it is necessary to detect and locate areas with streaks in a panoramic map, the panoramic map to be detected is first obtained, and a ground elevation model constructed based on the image that constructed the panoramic map is also obtained. This model is used as the digital surface model corresponding to the panoramic map. This digital surface model can be used to represent the height information of various planar features in the panoramic map.

[0045] For example, the panoramic map in this embodiment of the invention can be a panoramic map of a photovoltaic power station. A drone can inspect each photovoltaic module in the photovoltaic power station for which a panoramic map needs to be constructed, following a planned path. Visible light images are collected at each collection point along the planned path, and the geographical coordinates of the collection point are recorded. These geographical coordinates are then saved as the geographical coordinates of the center point of the visible light image collected at that point. After the data collection is completed, the visible light images containing the geographical coordinate information are input into panoramic map generation software to generate the panoramic map and ground elevation model. For example, the panoramic map generation software can be pix4dMapper; this embodiment of the invention does not limit this.

[0046] Furthermore, due to the diverse locations of photovoltaic power plants, existing technologies for rapid stitching of multiple consecutive photovoltaic module images still suffer from poor generalization. For example, when constructing panoramic maps for surface power plants and floating power plants, the visible light images collected in the power plants only contain the water surface and photovoltaic modules, resulting in high image similarity. When using panoramic map generation software as described in the above example to generate panoramic maps, some stuttering will still occur, severely affecting the quality of the resulting panoramic map stitching.

[0047] S102. Based on the digital land surface model, at least one component area image to be processed is determined from the panoramic map by segmentation.

[0048] In this embodiment, the image of the component area to be processed can be specifically understood as the image corresponding to the area where the photovoltaic module is located in the panoramic map.

[0049] Specifically, since the "patterning" phenomenon often occurs in photovoltaic (PV) module areas, when locating the target patterning area, the corresponding regions of each PV module in the panoramic image can be segmented first, and each region containing consecutive PV modules can be identified as the module region image to be processed. Furthermore, since there is a correspondence between the digital terrain model and the panoramic map, it can be used to represent the height information corresponding to each coordinate point in the panoramic map. Since the height of the areas where PV modules are located follows a fixed pattern, and the shapes of the areas where PV modules are located are regular, the regions containing the modules can be located based on the elevation information in the digital terrain model. Based on this location, at least one module region image to be processed can be segmented from the panoramic map.

[0050] S103. Extract features from the images of each component region to be processed, and determine the initial latte art region image based on the extracted features.

[0051] In this embodiment, the initial latte art area image can be specifically understood as the image in which the latte art phenomenon exists in the area image of each component to be processed.

[0052] Specifically, since not all component area images to be processed have the "scratching" phenomenon, feature extraction is first required for each component area image to obtain its corresponding image features. Since the image features of normally stitched component areas should be regularly distributed, and the image features of areas with "scratching" phenomenon should also be different from those of normally stitched areas due to stitching errors, feature comparison and recognition can be performed on each component area image after feature extraction. The component area images with the corresponding image features of "scratching" phenomenon are then determined as the initial "scratching" area images.

[0053] In this embodiment of the invention, the component area with the latte art phenomenon is first determined by image feature extraction, and then the latte art area is located in a smaller range. The method of secondary positioning is adopted to improve the accuracy of latte art area positioning.

[0054] S104. Input the initial latte art region image into the pre-built component segmentation neural network model, and determine the target latte art region in the initial latte art region image based on the output target segmentation image.

[0055] In this embodiment, the component segmentation neural network model can be specifically understood as a neural network model used to perform binary classification on the input image according to pre-trained rules, and then partition the different types of regions within it. The target segmented image can be specifically understood as the image output by the pre-built component segmentation neural network model after dividing the initial latte art region image. The target latte art region can be specifically understood as the component region in the initial latte art region image that specifically contains the latte art phenomenon.

[0056] Specifically, the initial latte art region image is input into a pre-trained component segmentation neural network model. The model performs binary classification on the initial latte art region image and outputs a target segmentation image containing component region location information. Shape features included in the location information of each component region are then extracted from the target segmentation image to determine whether the component corresponding to the location information has a latte art phenomenon. The bounding rectangles of components with latte art phenomena and interconnected components are defined as the target latte art region location boxes, and the regions located within the target latte art region location boxes are defined as the target latte art regions.

[0057] The technical solution of this embodiment involves acquiring a panoramic map of a photovoltaic power station and a digital surface model corresponding to the panoramic map; based on the digital surface model, segmenting and determining at least one component area image to be processed from the panoramic map; extracting features from each component area image to be processed, and determining an initial pattern area image based on the extracted features; inputting the initial pattern area image into a pre-constructed component segmentation neural network model, and determining the target pattern area in the initial pattern area image based on the output target segmentation image. By adopting the above technical solution, after obtaining the panoramic map and the corresponding digital surface model, information is first extracted and segmented from the panoramic map based on the digital surface model. The areas containing photovoltaic modules to be processed are extracted. Then, by extracting image features from each area to be processed, the initial images of areas with potential "streaking" (or "patterning") are identified. Combined with a pre-trained component segmentation neural network model, the initial images of the streaking areas are segmented into components, allowing for more refined segmentation. Based on the segmented target images, the location boxes of the target streaking areas with streaking are accurately determined. This solves the problem of accurately locating areas with streaking in existing panoramic electronic map stitching processes, improving the accuracy and precision of streaking area location in panoramic maps. This allows maintenance personnel to promptly re-acquire and mosaic the streaking areas, thereby improving the accuracy of component fault detection based on panoramic maps.

[0058] Example 2

[0059] Figure 2This is a flowchart of a panoramic map pattern detection method provided in Embodiment 2 of the present invention. The technical solution of the present invention is further optimized based on the above-mentioned optional technical solutions. The coordinate information of the component region to be processed in the panoramic map is determined by the elevation information, preset elevation threshold and preset boundary conditions in the digital surface model. At least one component region image to be processed is extracted from the panoramic map. Image feature extraction is performed on each component region image. Based on the extracted straight line related information and the preset first pattern judgment condition, the component region image with pattern phenomenon is determined as the initial pattern region image. The initial pattern region image is further input into the pre-constructed component segmentation neural network model to complete more accurate component region segmentation. The shape and area of ​​the component region are determined according to the connectivity of the component region. Then, each component region is judged according to the preset second pattern judgment condition to determine the target connected region image. The rectangle with the minimum bounding box of each target connected region image is determined as the target pattern region positioning box. The target pattern region in the initial pattern region image is determined based on the target pattern region positioning box. Meanwhile, Embodiment 2 of the present invention also provides a method for constructing and training a component segmentation neural network model, which combines image feature extraction method, neural network model and connected feature extraction method, sets different latte art judgment conditions to complete the accurate positioning of the latte art area in a progressive manner, improves the accuracy and precision of the latte art area positioning in the panoramic map, and enables maintenance personnel to perform image re-acquisition and mosaic repair of the latte art area in a timely manner.

[0060] like Figure 2 As shown, the panoramic map decal detection method provided in Embodiment 2 of the present invention specifically includes the following steps:

[0061] S201. Obtain a panoramic map of the photovoltaic power station and a digital land model corresponding to the panoramic map.

[0062] S202. Determine the first region set based on the elevation information in the digital surface model and the preset elevation threshold.

[0063] In this embodiment, the preset elevation threshold can be understood as a pre-set threshold used to determine the possible areas where photovoltaic modules may exist in the digital land model. It can be set according to the general height of photovoltaic modules, or it can be set adaptively according to the actual situation. This embodiment of the invention does not limit this.

[0064] Specifically, since the digital land model can be used to reflect the elevation information of all existing elements in its corresponding image, and the height of the photovoltaic modules located in the panoramic map should be approximately higher than the height of interfering factors such as shrubs, a preset elevation threshold can be set in advance. Areas in the digital land model with elevation information greater than the preset elevation threshold are identified as areas where photovoltaic modules may exist, and the set of the above areas is identified as the first area set.

[0065] S203. Based on the boundary characteristics of each first region in the first region set, the first region whose boundary characteristics meet the preset boundary conditions is determined as the component region to be processed.

[0066] In this embodiment, the preset boundary condition can be specifically understood as a determination condition used to determine whether a connected region is the region where the module is located. Optionally, the preset boundary condition can be a rectangular boundary of a closed region, or other characteristic conditions that can determine the region where the photovoltaic module is located; this embodiment of the invention does not limit this. The region of the module to be processed can be specifically understood as the region where the photovoltaic module where the streaking phenomenon needs to be determined is located.

[0067] For example, Figure 3 This is an example diagram of a first region set provided in Embodiment 2 of the present invention, as shown below. Figure 3 As shown, the first set of regions extracted based on the preset elevation threshold includes many irregular or discontinuous point-like regions. In a photovoltaic power station, photovoltaic modules should exist in a regular shape. By identifying and filtering the boundary features, the regions with regular shapes can be further refined to determine the areas where photovoltaic modules should be located.

[0068] Specifically, a continuous region in the first region set is defined as a first region. Boundary extraction algorithm is used to extract the boundary features of each first region. Then, the boundary features of each first region are compared with preset boundary conditions. When the preset boundary condition is that the boundary of the closed region is rectangular, the first regions in the first region set with rectangular boundary features can be selected and retained, while irregular first regions that do not conform to the rectangular features are deleted. Then, the region corresponding to the selected first region is determined as the component region to be processed. It can be considered that the component region to be processed corresponds to the region where the continuous photovoltaic modules of the photovoltaic power station are located.

[0069] For example, Figure 4 This is an example diagram of a component region to be processed provided in Embodiment 2 of the present invention, as shown below. Figure 4 As shown, this includes multiple component regions to be processed after filtering, which have been... Figure 3 The first region in the first region set shown is deleted if the boundary features of continuous regions do not conform to the rectangular feature.

[0070] In this embodiment of the invention, regions with both height and shape that conform to the requirements are determined from the digital surface model by sequentially using elevation information and boundary features. These regions are then used as potential photovoltaic module regions to be processed. This makes the features of the determined regions to be processed more similar to the features of the regions where the photovoltaic modules are located, increasing the likelihood of detecting the "streaking" phenomenon in the regions determined in the panoramic map based on the regions to be processed.

[0071] S204. Based on the coordinate information corresponding to each component region to be processed, extract the image of the component region to be processed corresponding to each component region from the panoramic map.

[0072] Specifically, since there is a correspondence between the digital surface model and the panoramic map, the coordinates in the digital surface model also correspond to the coordinates in the panoramic map. After determining the coordinate information corresponding to each component area to be processed, the corresponding coordinates in the panoramic map can be determined based on the coordinate information, and the image within the corresponding coordinate range can be determined as the image of the component area to be processed corresponding to each component area.

[0073] In this embodiment of the invention, the combination of digital land surface model and panoramic map is used to segment the image of the component area to be processed corresponding to the area where photovoltaic modules may exist from the panoramic map. Compared with the direct extraction and identification of image features based on the panoramic map, this method fully considers the features of the component area in terms of shape and elevation, thereby improving the accuracy of determining the image of the component area to be processed.

[0074] S205. Perform grayscale and edge processing on the images of each component region to be processed, and determine the edge images of the component regions to be processed.

[0075] Specifically, since the image of the component area to be processed extracted from the panoramic map is a visible light image, in order to determine whether there is a pattern, the straight line information needs to be extracted based on the characteristics of the photovoltaic module. Therefore, it is first grayscale processed, and then Canny edge processing is performed to obtain the edge image of the component area to be processed after edge processing.

[0076] S206. Extract the set of straight lines from the edge images of each component region to be processed.

[0077] For example, the Hough line transform method can be used to search in the boundary images of each component region to be processed to determine all possible straight lines, and the set of all straight lines is determined as the set of straight lines corresponding to the edge image of the component region to be processed.

[0078] S207. The image of the component region to be processed corresponding to the set of straight lines that meet the first preset latte art judgment condition is determined as the initial latte art region image.

[0079] In this embodiment, the first preset streaking determination condition can be specifically understood as a condition determined based on the straight line characteristics in the photovoltaic module and the straight line characteristics that may exist in the photovoltaic module when streaking occurs, used to determine whether streaking occurs in a large area of ​​the module.

[0080] Specifically, the corresponding line features are extracted from the set of lines, and the line features are compared with the first preset latte art judgment condition. If they meet the first preset latte art judgment condition, it can be preliminarily determined that there is a latte art phenomenon in the image of the component area to be processed corresponding to the set of lines, and it is used as the initial latte art area image. All sets of lines are judged in turn until all images of the component area to be processed are judged.

[0081] Furthermore, for each set of lines, the set of lines is divided into a long line subset and a short line subset according to a preset length threshold; if the first average length of the long line subset, the second average length of the short line subset, the number of the first element of the long line subset, and the number of the second element of the short line subset all satisfy the first preset latte art determination condition, then the image of the component region to be processed corresponding to the set of lines is determined as the initial latte art region image.

[0082] The first preset criteria for determining the latte art are that the average length of the first element is greater than the average length of the second element, the number of the first element is greater than the first quantity threshold, and the number of the second element is greater than the second quantity threshold.

[0083] In this embodiment, the preset length threshold can be specifically understood as a length set based on the length characteristics of continuous straight lines in the photovoltaic module, used to determine whether there is a "streaking" phenomenon in the same photovoltaic module. The first preset streaking determination condition can be specifically understood as a determination condition constructed based on the average length of long straight lines and the number of long and short straight lines, used to determine whether there is a streaking phenomenon in the image of the component region to be processed corresponding to the set of straight lines. It should be clarified that if there is no streaking phenomenon, there should be continuous long straight lines in the streaking area, and the number of long straight lines should reach a certain threshold. When there is a streaking phenomenon, the streaking phenomenon caused by splicing will break the continuous long straight lines into multiple short straight lines. Therefore, the set of straight lines can be divided according to the determined length threshold, and the average length and number of long and short straight lines in the set can be used to determine whether the image of the component region to be processed corresponding to the set of straight lines is the initial streaking area image including the streaking phenomenon.

[0084] For example, Figure 5 This is an example diagram of a long straight line subset provided in Embodiment 2 of the present invention. Figure 6 This is an example diagram of a short straight line subset provided in Embodiment 2 of the present invention. Figure 5 and Figure 6As shown, the overall rectangular area is used to represent the image of the component region to be processed. After identifying all straight lines through Hough transform, the lengths of all straight lines are calculated based on the coordinate information in the image of the component region to be processed. Then, according to a preset length threshold, lines greater than the preset length threshold are identified as long straight lines, and the set of all long straight lines is identified as a subset of long straight lines, such as... Figure 5 As shown, the set of all black lines is the long straight line subset; lines less than or equal to the preset length threshold are defined as short straight lines, and the set of all short straight lines is defined as the short straight line subset, as shown below. Figure 6 As shown, the set of all black lines is the subset of short straight lines.

[0085] Specifically, for a set of lines, it is divided into a long line subset L and a short line subset S according to a preset length threshold, and the average length of the long line subset L is determined as the first average length L. avg Let the number of long lines in the long line subset L be the first element number N, and let the average length of the short line subset S be the second average length S. avg Let the number of short lines in the subset S be defined as the number of the second element M. Assuming the first preset multiple is 5, the first quantity threshold is 5, and the second quantity threshold is 5, then when L avg S avg N and M satisfy the condition {L avg >5×S avg When N > 5 and M > 5, the image of the component region to be processed corresponding to the set of straight lines can be considered as the initial latte art region image. It should be clarified that the first preset multiple, the first quantity threshold, and the second quantity threshold can be set according to actual conditions. This embodiment of the invention only uses the above-mentioned value range as an example and does not impose specific limitations.

[0086] Furthermore, Figure 7 This is a flowchart illustrating a training method for a component segmentation neural network model provided in Embodiment 2 of the present invention, as shown below. Figure 7 As shown, the specific steps include the following:

[0087] S301. Obtain the training set of the component segmentation model.

[0088] The training set of the component segmentation model includes a set of real images and a set of labeled images corresponding to the real images. The set of labeled images includes pixel category information corresponding to the real images.

[0089] In this embodiment, the component segmentation model training set can be specifically understood as the set of training objects determined based on real images, used to input into the untrained component segmentation neural network model for training. Further, since the component segmentation neural network model in this application is a model used to segment components based on a binary classification method according to the input model, the input component segmentation training set should include a set of real images consisting of visible light images corresponding to the component regions, used for segmentation training, and a set of labeled images corresponding to the real image set after manually labeling the pixel categories of each component.

[0090] S302. Input the real image set into the initial component segmentation neural network model to obtain the intermediate segmented image set output by the initial component segmentation neural network model.

[0091] In this embodiment, the initial component segmentation neural network model can be understood as the component segmentation neural network model before training. The neural network layer structure is completely consistent with the component segmentation neural network model, and can be viewed as a neural network model composed of an upsampling module and a downsampling module, but the weight parameters of each neural network layer have not yet been adjusted. The intermediate segmented image set can be understood as the result of the initial component segmentation neural network model before training, which segments the real images based on the input real image set.

[0092] Specifically, the real image set is input into the initial component segmentation neural network model for training. During the training process, the different intermediate segmentation image results output by the initial component segmentation neural network model for different real images can be extracted, and the set of each intermediate segmentation image is determined as the intermediate segmentation image set.

[0093] S303. Construct a loss function based on the intermediate segmented image set and the calibration image set.

[0094] Specifically, since the intermediate segmentation image set and the calibration image set contain the same number of images and have a corresponding relationship, the calibration information contained in the calibrated calibration image set and its corresponding intermediate segmentation image can be compared, and the corresponding loss function can be determined based on the comparison result.

[0095] Optionally, since the component segmentation is a binary classification problem, the loss function type is selected in this embodiment of the invention. The logistic loss function can be used as the objective function to measure the deviation between the predicted value and the actual value. The loss function can be expressed by the following formula:

[0096]

[0097] Where i represents the number of intermediate segmented images in the intermediate segmented image set, L represents the loss value, y represents the predicted value, y' represents the actual value, and m represents the number of categories.

[0098] S304. Train the initial component segmentation neural network model using the loss function until the preset convergence condition is met, and obtain the trained component segmentation neural network model.

[0099] In this embodiment, the preset convergence condition can be specifically understood as a condition used to determine whether the initially trained component segmentation neural network model has entered a convergence state. Optionally, the preset convergence condition may include the change in weight parameters between two iterations of model training being less than a preset parameter change threshold, or the iteration exceeding a set maximum number of iterations, or all samples in the component segmentation model training set having been trained, etc., and this embodiment of the invention does not limit this.

[0100] Specifically, backpropagation is performed on the initial component segmentation neural network model based on the loss function, so that the weight parameters in each neural network layer used to form the initial component segmentation neural network model can be adjusted according to the loss function until the preset convergence condition is met, and the trained initial component segmentation neural network model is determined as the component segmentation neural network model.

[0101] S208. Input the initial latte art region image into the downsampling module of the pre-built component segmentation neural network model to determine at least two feature images of different scales.

[0102] For example, Figure 8 This is an example diagram of the network structure of a component segmentation neural network model provided in Embodiment 2 of the present invention, as shown below. Figure 8 As shown, it may include a downsampling module as an encoding structure and an upsampling module as a decoding structure. For example, the downsampling module consists of two 3*3 convolutional layers plus a 2*2 pooling layer, and the upsampling module consists of a deconvolutional layer, a feature concatenation layer, and two 3*3 convolutional layers.

[0103] Following the example above, assuming the initial latte art area image is 224*224 in size, then input it into... Figure 8 In the downsampling module of the component segmentation neural network model shown, it can be extracted into feature images of four different scales: 112*112, 56*56, 28*28, and 14*14.

[0104] S209. Input each feature image into the upsampling module of the component segmentation neural network model to determine the output feature image.

[0105] Following the example above, input feature images of four different scales—112*112, 56*56, 28*28, and 14*14—into a program like this: Figure 8 The component shown decodes the feature image in the upsampling module of the segmentation neural network model. First, it deconvolves the 14*14 feature image to obtain a 28*28 feature image. Then, it concatenates the 28*28 feature image with the 28*28 feature image from the previous downsampling module. The concatenated feature image is then convolved and deconvolved to obtain a 56*56 feature image. This 56*56 feature image is then concatenated with the 56*56 feature image from the previous downsampling module. Through convolution and deconvolution, after four deconvolution operations, a 224*224 segmentation image with the same size as the initial latte art region image is obtained. The output image is then determined as the output feature image corresponding to the initial latte art region image.

[0106] S210. The result of the output feature image through the loss layer is determined as the target segmentation image.

[0107] Specifically, the output feature image is passed to the logistic loss layer, and the output of the logistic loss layer is used to determine the target segmentation image. For example, Figure 9 This is an example image of a target segmentation image provided in Embodiment 2 of the present invention, wherein the white blocks represent photovoltaic modules.

[0108] S211. Determine the set of connected component images in the output target segmentation image based on connectivity.

[0109] Specifically, since correctly spliced ​​photovoltaic modules should be distributed in a regular rectangular pattern, in order to determine whether there is a "streaking" phenomenon in the target segmentation image obtained after module segmentation, it is necessary to determine the connected regions of the image through connectivity. Theoretically, images belonging to the same connected region can be considered as the same photovoltaic module, and the set of the determined connected regions is defined as the connected region image set.

[0110] S212. Based on the average image area corresponding to the set of connected region images, the width of each connected region image, the height of each connected region image, and the second preset latte art judgment condition, determine at least one target connected region image.

[0111] The second preset criteria for determining the pattern is that the area of ​​the connected region image is equal to the product of the width and height of the connected region image, and the area of ​​the connected region image is greater than the average image area of ​​the second preset multiple.

[0112] In this embodiment, the second preset pattern determination condition can be specifically understood as a condition based on the shape characteristics of the photovoltaic module to determine whether each photovoltaic module is a regular image within a small range, and then to determine whether there is a pattern phenomenon at its corresponding position.

[0113] Specifically, based on the set of connected region images, the average image area of ​​each connected region image, as well as the width and height of each connected region image, are determined. If a connected region image is a regular rectangle, its area should be equal to the product of its length and width, and its area should be at least less than twice the average image area. Therefore, in this embodiment of the invention, connected regions whose area is equal to the product of the width and height of the connected region image, and whose area is greater than the average image area by a second preset multiple, are determined as regions with a "streaking" phenomenon. These regions with a "streaking" phenomenon are then determined as target connected region images.

[0114] Following the example above, with Figure 9 Taking the example image of the target segmentation image as an example, the connected regions are determined based on connectivity, and the coordinates of the connected regions represented by white are obtained. Then, the average area of ​​the white blocks is calculated as the average image area A. avg Assuming the number of connected regions in the target segmented image is Z, W i H represents the width of the i-th connected region. i Let represent the height of the i-th connected region. Then, the average image area can be expressed by the following formula:

[0115]

[0116] Furthermore, assuming the second preset multiple is 2, S i Let A represent the area of ​​the i-th connected region. Then, when the i-th connected region satisfies the condition {A}... i =W i ×H i And A i >2A avg When}, the i-th connected region can be considered as the target connected region image.

[0117] S213. Determine the smallest bounding rectangle of each target connected region image as the target latte art region location box, and determine the target latte art region in the initial latte art region image based on the target latte art region location box.

[0118] In this embodiment, the target latte art area positioning box can be specifically understood as the smallest outer rectangle of the component area in the initial latte art area image where the latte art phenomenon specifically exists.

[0119] Specifically, since the shapes corresponding to each target connected region are not regular, to clearly indicate their position in the panoramic map, the coordinates of the smallest bounding rectangle of the target connected region can be determined based on the coordinates of the identified target connected region. This rectangle is then used as the target latte art region positioning box to detect and locate the latte art region in the panoramic map. The area within the target latte art region positioning box is defined as the target latte art region in the initial latte art region image. For example, Figure 10 This is an example diagram of a target latte art area positioning frame provided in Embodiment 2 of the present invention, wherein the rectangular frame formed by black lines is the target latte art area positioning frame.

[0120] The technical solution of this embodiment first extracts areas where photovoltaic modules may exist based on elevation information from a digital surface model according to a preset elevation threshold. Then, it identifies boundary conditions for the extracted areas that meet the elevation conditions, segmenting at least one module area image to be processed from the panoramic map, making the determined module area images more accurate. Then, based on feature extraction of each module area image, and based on the extracted straight line related information and a first preset pattern determination condition, it determines the initial pattern area image that may have patterning from multiple module area images. Then, it inputs the initial pattern area image into a pre-constructed module segmentation neural network model for precise module segmentation. Based on the connectivity of the segmented module areas, it determines their corresponding shape and area. Then, based on a pre-set second preset pattern determination condition, it determines each module area to determine the target connected region image. By using two different directions and different determination conditions for patterning, it performs two levels of precision determination on the areas with patterning, from coarse to fine, making the determined target connected region image position more accurate. This makes the target pattern area range defined by the target pattern area positioning box determined based on the minimum bounding rectangle of the target connected region image more accurate. Because this invention combines image feature extraction methods, neural network models, and connected feature extraction methods, and sets different criteria for determining the pattern to achieve precise localization of the pattern area in a progressive manner, it improves the accuracy and precision of the pattern area localization in the panoramic map, thereby enabling maintenance personnel to promptly re-acquire and mosaic the pattern area.

[0121] Example 3

[0122] Figure 11 This is a schematic diagram of a panoramic map latte art detection device provided in Embodiment 3 of the present invention. The panoramic map latte art detection device includes: a map model acquisition module 41, a to-be-processed image determination module 42, an initial image determination module 43, and a latte art area determination module 44.

[0123] The system includes a map model acquisition module 41, which acquires a panoramic map of the photovoltaic power station and a digital surface model corresponding to the panoramic map; an image determination module 42, which determines at least one component region image to be processed from the panoramic map based on the digital surface model; an initial image determination module 43, which extracts features from each component region image to be processed and determines an initial patterned region image based on the extracted features; and a patterned region determination module 44, which inputs the initial patterned region image into a pre-built component segmentation neural network model and determines the target patterned region in the initial patterned region image based on the output target segmentation image.

[0124] The technical solution of this embodiment, after acquiring the panoramic map and the corresponding digital surface model, firstly extracts and segments the panoramic map based on the digital surface model, extracting the areas containing photovoltaic modules to be processed. Then, by extracting image features from each area to be processed, the initial image of the area with potential "streaking" (or "bleeding") is identified. A pre-trained component segmentation neural network model is then used to segment the initial image of the "streaking" area, allowing for more refined segmentation. Based on the segmented target image, the target "streaking" area with the "streaking" phenomenon is accurately determined. This solves the problem of accurately locating areas with "streaking" phenomena during existing panoramic electronic map stitching processes, improving the accuracy and precision of locating "streaking" areas in the panoramic map. This enables maintenance personnel to promptly re-acquire and mosaic the "streaking" areas, thereby improving the accuracy of component fault detection based on the panoramic map.

[0125] Optionally, the image-to-process determination module 42 includes:

[0126] The first set determination unit is used to determine the first region set based on the elevation information in the digital surface model and a preset elevation threshold.

[0127] The region to be processed determination unit is used to determine the first region whose boundary features meet the preset boundary conditions as the component region to be processed based on the boundary features of each first region in the first region set.

[0128] The image to be processed determination unit is used to extract the image of the component region corresponding to each component region from the panoramic map based on the coordinate information corresponding to each component region to be processed.

[0129] Optionally, the initial image determination module 43 includes:

[0130] An edge image determination unit is used to perform grayscale and edge processing on the images of each of the regions to be processed, and to determine the edge images of the regions to be processed.

[0131] A line set determination unit is used to extract the line set from the edge image of each component region to be processed;

[0132] The initial image determination unit is used to determine the image of the component region to be processed corresponding to the set of straight lines that meet the first preset latte art determination condition as the initial latte art region image.

[0133] Furthermore, the initial image determination unit is specifically used for:

[0134] For each set of lines, the set of lines is divided into a long line subset and a short line subset according to a preset length threshold;

[0135] If the first average length of the long straight line subset, the second average length of the short straight line subset, the number of the first element of the long straight line subset, and the number of the second element of the short straight line subset all satisfy the first preset latte art determination condition, then the image of the component region to be processed corresponding to the straight line set is determined as the initial latte art region image.

[0136] The first preset latte art determination condition is that the average length of the first element is greater than the average length of the second element by a first preset multiple, the number of the first element is greater than the first quantity threshold, and the number of the second element is greater than the second quantity threshold.

[0137] Optionally, the latte art area determination module 44 includes:

[0138] The target image determination unit is used to input the initial latte art region image into the downsampling module of the pre-built component segmentation neural network model to determine at least two feature images of different scales; input each of the feature images into the upsampling module of the component segmentation neural network model to determine the output feature image; and determine the result of the output feature image through the loss layer as the target segmentation image.

[0139] The latte art region determination unit is used to determine a set of connected region images in the output target segmentation image based on connectivity; determine at least one target connected region image based on the average image area corresponding to the set of connected region images, the width of each connected region image, the height of each connected region image, and a second preset latte art judgment condition; determine the smallest rectangle bounded to each of the target connected region images as the target latte art region positioning box, and determine the target latte art region in the initial latte art region image based on the target latte art region positioning box.

[0140] The second preset criteria for determining the latte art is that the area of ​​the connected region image is equal to the product of the width and the height of the connected region image, and the area of ​​the connected region image is greater than the average image area by a second preset multiple.

[0141] Furthermore, the construction of the component segmentation neural network model includes:

[0142] Obtain a training set for the component segmentation model, which includes a set of real images and a set of labeled images corresponding to the set of real images. The set of labeled images includes pixel category information corresponding to the set of real images.

[0143] The real image set is input into the initial component segmentation neural network model to obtain the intermediate segmented image set output by the initial component segmentation neural network model;

[0144] A loss function is constructed based on the intermediate segmented image set and the calibration image set;

[0145] The initial component segmentation neural network model is trained using the loss function until a preset convergence condition is met, resulting in a trained component segmentation neural network model.

[0146] The loss function is the logistic loss function, which can be expressed by the following formula:

[0147]

[0148] Where i represents the number of intermediate segmented images in the intermediate segmented image set, L represents the loss value, y represents the predicted value, y' represents the actual value, and m represents the number of categories.

[0149] The panoramic map decal detection device provided in this embodiment of the invention can execute the panoramic map decal detection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0150] Example 4

[0151] Figure 12 This is a schematic diagram of the structure of a panoramic map decal detection device according to Embodiment 4 of the present invention. The panoramic map decal detection device 50 can be an electronic device, intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0152] like Figure 12As shown, the panoramic map latte art detection device 50 includes at least one processor 51 and a memory, such as a read-only memory (ROM) 52 and a random access memory (RAM) 53, communicatively connected to the at least one processor 51. The memory stores computer programs executable by the at least one processor. The processor 51 can perform various appropriate actions and processes based on the computer program stored in the ROM 52 or loaded from the storage unit 58 into the RAM 53. The RAM 53 can also store various programs and data required for the operation of the panoramic map latte art detection device 50. The processor 51, ROM 52, and RAM 53 are interconnected via a bus 54. An input / output (I / O) interface 55 is also connected to the bus 54.

[0153] Multiple components in the panoramic map decal detection device 50 are connected to the I / O interface 55, including: an input unit 56, such as a keyboard, mouse, etc.; an output unit 57, such as various types of displays, speakers, etc.; a storage unit 58, such as a disk, optical disk, etc.; and a communication unit 59, such as a network card, modem, wireless transceiver, etc. The communication unit 59 allows the panoramic map decal detection device 50 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0154] Processor 51 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 51 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 51 performs the various methods and processes described above, such as the panoramic map latte art detection method.

[0155] In some embodiments, the panoramic map latte art detection method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 58. In some embodiments, part or all of the computer program may be loaded and / or installed on the panoramic map latte art detection device 50 via ROM 52 and / or communication unit 59. When the computer program is loaded into RAM 53 and executed by processor 51, one or more steps of the panoramic map latte art detection method described above may be performed. Alternatively, in other embodiments, processor 51 may be configured to perform the panoramic map latte art detection method by any other suitable means (e.g., by means of firmware).

[0156] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0157] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0158] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0159] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0160] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0161] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0162] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0163] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for detecting decals on panoramic maps, characterized in that, include: Obtain a panoramic map of the photovoltaic power station and a digital land surface model corresponding to the panoramic map; Based on the digital land surface model, at least one component area image to be processed is segmented and determined from the panoramic map; Feature extraction is performed on the images of each component region to be processed, and the initial latte art region image is determined based on the extracted features; The initial latte art region image is input into a pre-built component segmentation neural network model, and the target latte art region in the initial latte art region image is determined based on the output target segmentation image; The step of extracting features from the images of each component region to be processed, and determining the initial latte art region image based on the extracted features, includes: Image features are extracted from the images of each component region to be processed. Based on the extracted straight line features and the first preset latte art determination condition, the images of the component regions to be processed that exhibit latte art are determined as the initial latte art region images. The first preset streaking determination condition is determined based on the straight line characteristics in the photovoltaic module and the straight line characteristics present in the photovoltaic module when streaking occurs, and is used to determine whether streaking occurs in the module area. The step of determining the target latte art region in the initial latte art region image based on the output target segmentation image includes: Determine the set of connected component images in the output target segmentation image based on connectivity; Based on the average image area corresponding to the set of connected region images, the width of each connected region image, the height of each connected region image, and the second preset latte art determination condition, at least one target connected region image is determined. The smallest bounding rectangle circumscribed in each of the target connected region images is determined as the target latte art region location box, and the target latte art region in the initial latte art region image is determined based on the target latte art region location box; The second preset criteria for determining the latte art is that the area of ​​the connected region image is equal to the product of the width and the height of the connected region image, and the area of ​​the connected region image is greater than the average image area by a second preset multiple.

2. The method according to claim 1, characterized in that, The step of segmenting and determining at least one component region image from the panoramic map based on the digital land surface model includes: Based on the elevation information in the digital surface model and the preset elevation threshold, a first set of regions is determined; Based on the boundary characteristics of each first region in the first region set, the first region whose boundary characteristics meet the preset boundary conditions is determined as the component region to be processed; Based on the coordinate information corresponding to each of the component regions to be processed, the images of the component regions to be processed corresponding to each of the component regions to be processed are extracted from the panoramic map.

3. The method according to claim 1, characterized in that, The step of extracting features from the images of each of the components to be processed, and determining the initial latte art region image based on the extracted features, includes: The images of each component region to be processed are subjected to grayscale and edge processing to determine the edge images of the component regions to be processed; Extract the set of straight lines from the edge images of each component region to be processed; The image of the component region to be processed corresponding to the set of straight lines that meet the first preset latte art judgment condition is determined as the initial latte art region image.

4. The method according to claim 3, characterized in that, The step of determining the image of the component region to be processed corresponding to the set of straight lines that meet the first preset latte art determination condition as the initial latte art region image includes: For each set of lines, the set of lines is divided into a long line subset and a short line subset according to a preset length threshold; If the first average length of the long straight line subset, the second average length of the short straight line subset, the first number of elements of the long straight line subset, and the second number of elements of the short straight line subset all satisfy the first preset latte art determination condition, then the image of the component region to be processed corresponding to the straight line set is determined as the initial latte art region image.

5. The method according to claim 4, characterized in that, The first preset latte art determination condition is that the average length of the first element is greater than the average length of the second element by a first preset multiple, the number of the first element is greater than the first quantity threshold, and the number of the second element is greater than the second quantity threshold.

6. The method according to claim 1, characterized in that, The step of inputting the initial latte art region image into a pre-built component segmentation neural network model includes: The initial latte art region image is input into the downsampling module of the pre-built component segmentation neural network model to determine at least two feature images of different scales; Each of the aforementioned feature images is input into the upsampling module of the component segmentation neural network model to determine the output feature image; The result of passing the output feature image through the loss layer is determined as the target segmentation image.

7. The method according to claim 1, characterized in that, The construction of the component segmentation neural network model includes: Obtain a training set for the component segmentation model, which includes a set of real images and a set of labeled images corresponding to the set of real images. The set of labeled images includes pixel category information corresponding to the set of real images. The real image set is input into the initial component segmentation neural network model to obtain the intermediate segmented image set output by the initial component segmentation neural network model; A loss function is constructed based on the intermediate segmented image set and the calibration image set; The initial component segmentation neural network model is trained using the loss function until a preset convergence condition is met, resulting in a trained component segmentation neural network model.

8. A panoramic map decal detection device, characterized in that, include: The map model acquisition module is used to acquire a panoramic map of the photovoltaic power station and a digital surface model corresponding to the panoramic map. The image to be processed determination module is used to determine at least one component area image to be processed from the panoramic map based on the digital land surface model. The initial image determination module is used to extract features from the images of the regions of each component to be processed, and to determine the initial latte art region image based on the extracted features. The latte art region determination module is used to input the initial latte art region image into a pre-built component segmentation neural network model, and determine the target latte art region in the initial latte art region image based on the output target segmentation image; The initial image determination module is specifically used for: Image features are extracted from the images of each component region to be processed. Based on the extracted straight line features and the first preset latte art determination condition, the images of the component regions to be processed that exhibit latte art are determined as the initial latte art region images. The first preset streaking determination condition is determined based on the straight line characteristics in the photovoltaic module and the straight line characteristics present in the photovoltaic module when streaking occurs, and is used to determine whether streaking occurs in the module area. The latte art area determination module includes: The latte art region determination unit is used to determine a set of connected region images in the output target segmentation image based on connectivity; determine at least one target connected region image based on the average image area corresponding to the set of connected region images, the width of each connected region image, the height of each connected region image, and a second preset latte art judgment condition; determine the smallest rectangle bounded to each of the target connected region images as the target latte art region positioning box, and determine the target latte art region in the initial latte art region image based on the target latte art region positioning box; The second preset criteria for determining the latte art is that the area of ​​the connected region image is equal to the product of the width and the height of the connected region image, and the area of ​​the connected region image is greater than the average image area by a second preset multiple.

9. A panoramic map decal detection device, characterized in that, The panoramic map decal detection device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the panoramic map latte art detection method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the panoramic map decal detection method according to any one of claims 1-7.

Images