Photovoltaic equipment ID matching method and system based on anchor point affine

Abstract

This paper proposes a photovoltaic device ID matching method and system based on anchor point affine. This method utilizes a neural network model to adaptively learn complex spatial mapping relationships, effectively addressing the distortion problem caused by varying viewing angles in large-scale photovoltaic arrays. Furthermore, this method combines a mapping point with a matching mechanism of adjacent feature points to select feature points that match the mapping point. Based on the feature point matching results, the Hungarian algorithm is used to further calibrate the matching relationship between the mapping point and its matching feature points, further improving the accuracy of photovoltaic device ID matching results.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and more particularly to a photovoltaic equipment ID matching method and system based on anchor point affine. Background Art

[0002] During the photovoltaic equipment installation project, it is necessary to match the photovoltaic equipment at the installation site with the photovoltaic equipment ID marked in the completion drawing one by one, so that the installation location of the photovoltaic equipment can be verified and the subsequent maintenance and management of the photovoltaic equipment can be facilitated.

[0003] Currently, the industry generally uses aerial photography to capture aerial images of photovoltaic equipment at installation sites. Image recognition technology is then used to match these images with the equipment IDs in as-built drawings. However, due to the limited viewing angle of aerial photography, this method often produces distorted and deformed images, resulting in low accuracy in identifying equipment locations and inaccurate ID matching. Summary of the Invention

[0004] Purpose of the invention: The present invention aims to overcome the defects of the prior art and provide a photovoltaic device ID matching method and system based on anchor point affine, which can improve the accuracy of photovoltaic device ID matching.

[0005] Summary of the invention: To achieve the above objectives, the present invention proposes the following technical solutions:

[0006] In a first aspect, a photovoltaic device ID matching method based on anchor point affine is provided, the method comprising:

[0007] Obtaining the aerial image of the photovoltaic array to be matched and the corresponding as-built image;

[0008] Extracting feature points of the photovoltaic equipment in the photovoltaic array aerial image and the photovoltaic equipment in the as-built image, and determining the coordinates of the feature points;

[0009] Using a pre-trained coordinate mapping model, the feature points in the as-built image are mapped to the photovoltaic array aerial image to obtain the coordinates of the mapping points;

[0010] In the photovoltaic array aerial image, performing feature matching between the mapping point and a plurality of adjacent feature points of the mapping point;

[0011] Taking the set of mapping points as a first point set, taking the feature points matched with the mapping points as a second point set, and calculating the optimal matching result between the first point set and the second point set using the Hungarian algorithm;

[0012] Based on the optimal matching result, the device ID corresponding to the mapping point is matched to the corresponding feature point in the photovoltaic array aerial image.

[0013] As an optional implementation of the method of the first aspect, the coordinate mapping model includes a first model based on a neural network model and a second model based on an interpolator;

[0014] Using a pre-trained coordinate mapping model, the feature points in the as-built image are mapped to the photovoltaic array aerial image to obtain the coordinates of the mapping points, specifically including:

[0015] Inputting the feature points in the as-built image into the first model to obtain a first coordinate mapping result;

[0016] Inputting the feature points in the as-built image into the second model to obtain a second coordinate mapping result;

[0017] A weighted combination is performed on the first coordinate mapping result and the second coordinate mapping result to obtain the coordinates of the mapping point.

[0018] Specifically, the method further includes:

[0019] Acquire a photovoltaic array aerial sample image and an as-built sample image corresponding to the photovoltaic array aerial sample image;

[0020] Selecting multiple sets of feature point sets in the as-built sample image to form feature point samples, and determining coordinates of target feature points corresponding to the feature point samples in the photovoltaic array aerial sample image;

[0021] Inputting the feature point samples into the coordinate mapping model to obtain predicted mapping point coordinates;

[0022] A loss function is constructed based on the predicted mapping point coordinates and the coordinates of the target feature points, and the coordinate mapping model is updated using the loss function until the coordinate mapping model that meets the requirements is obtained.

[0023] As an optional implementation manner of the method of the first aspect, the method further includes:

[0024] Perform OCR recognition on the as-built image to extract the ID of the photovoltaic equipment in the as-built image.

[0025] As an optional implementation manner of the method of the first aspect, in the photovoltaic array aerial image, feature matching is performed on the mapping point with several adjacent feature points of the mapping point, specifically including:

[0026] Calculate the Euclidean distance cost Distance_Cost between the mapping point and the adjacent feature point;

[0027] Calculate the relative position relationship cost Position_Cost between the mapping point and the adjacent feature point:

[0028] ;

[0029] Wherein, p represents the original feature point corresponding to the mapping point in the as-built image, nᵢ represents the i-th adjacent feature point around the original feature point p in the as-built image, i=1,2,…,k, k is a positive integer, d(p,nᵢ) represents the distance between the original feature point p and the adjacent feature point nᵢ; p' represents the adjacent feature point of the mapping point in the PV array aerial image, nᵢ' represents the i-th adjacent feature point around the adjacent feature point p', i=1,2,…,k, d'(p',nᵢ') represents the distance between the adjacent feature point p' and its adjacent feature point nᵢ';

[0030] Calculate the consistency cost Anchor_Cost of the anchor point between the mapping point and the adjacent feature point:

[0031] ;

[0032] in, represents the jth anchor point in the as-built image, Indicates the original feature point p and the anchor point in the completed image The distance ratio between them, j=1,2,…,m, m is a positive integer; Indicates that in the aerial image of the photovoltaic array The corresponding anchor point, Represents the adjacent feature point p' and the anchor point The distance ratio between

[0033] Calculate the matching cost:

[0034] ;

[0035] in, and represents the weight coefficient;

[0036] Based on the matching cost, feature matching is performed on the mapping point and a number of adjacent feature points of the mapping point.

[0037] In a second aspect, a photovoltaic device ID matching system based on anchor point affine is provided, the system comprising:

[0038] A data acquisition module is used to obtain the aerial image of the photovoltaic array to be matched and the corresponding as-built image;

[0039] a data preprocessing module, configured to extract feature points of the photovoltaic devices in the photovoltaic array aerial image and the photovoltaic devices in the as-built image, and determine the coordinates of the feature points;

[0040] A coordinate mapping module is used to map the feature points in the as-built image to the photovoltaic array aerial image using a pre-trained coordinate mapping model to obtain the coordinates of the mapping points;

[0041] The matching module is configured to perform feature matching between the mapping point and several adjacent feature points of the mapping point in the aerial image of the photovoltaic array; use the set of mapping points as a first point set, use the feature points matched with the mapping points as a second point set, and calculate an optimal matching result between the first point set and the second point set using the Hungarian algorithm; and match the device ID corresponding to the mapping point to the corresponding feature point in the aerial image of the photovoltaic array based on the optimal matching result.

[0042] As an optional implementation of the system of the second aspect, the coordinate mapping model includes a first model based on a neural network model and a second model based on an interpolator; the coordinate mapping module is specifically configured to:

[0043] Inputting the feature points in the as-built image into the first model to obtain a first coordinate mapping result;

[0044] Inputting the feature points in the as-built image into the second model to obtain a second coordinate mapping result;

[0045] A weighted combination is performed on the first coordinate mapping result and the second coordinate mapping result to obtain the coordinates of the mapping point.

[0046] Specifically, the system further includes a training module, which is used to:

[0047] Acquire a photovoltaic array aerial sample image and an as-built sample image corresponding to the photovoltaic array aerial sample image;

[0048] Selecting multiple sets of feature point sets in the as-built sample image to form feature point samples, and determining coordinates of target feature points corresponding to the feature point samples in the photovoltaic array aerial sample image;

[0049] Inputting the feature point samples into the coordinate mapping model to obtain predicted mapping point coordinates;

[0050] A loss function is constructed based on the predicted mapping point coordinates and the coordinates of the target feature points, and the coordinate mapping model is updated using the loss function until the coordinate mapping model that meets the requirements is obtained.

[0051] As an optional implementation manner of the system according to the second aspect, the data preprocessing module is further configured to:

[0052] Perform OCR recognition on the as-built image to extract the ID of the photovoltaic equipment in the as-built image.

[0053] As an optional implementation manner of the system according to the second aspect, the matching module is specifically configured to:

[0054] Calculate the Euclidean distance cost Distance_Cost between the mapping point and the adjacent feature point;

[0055] Calculate the relative position relationship cost Position_Cost between the mapping point and the adjacent feature point:

[0056] ;

[0057] Wherein, p represents the original feature point corresponding to the mapping point in the as-built image, nᵢ represents the i-th adjacent feature point around the original feature point p in the as-built image, i=1,2,…,k, k is a positive integer, d(p,nᵢ) represents the distance between the original feature point p and the adjacent feature point nᵢ; p' represents the adjacent feature point of the mapping point in the PV array aerial image, nᵢ' represents the i-th adjacent feature point around the adjacent feature point p', i=1,2,…,k, d'(p',nᵢ') represents the distance between the adjacent feature point p' and its adjacent feature point nᵢ';

[0058] Calculate the consistency cost Anchor_Cost of the anchor point between the mapping point and the adjacent feature point:

[0059] ;

[0060] in, represents the jth anchor point in the as-built image, Indicates the original feature point p and the anchor point in the completed image The distance ratio between them, j=1,2,…,m, m is a positive integer; Indicates that in the aerial image of the photovoltaic array The corresponding anchor point, Represents the adjacent feature point p' and the anchor point The distance ratio between

[0061] Calculate the matching cost:

[0062] ;

[0063] in, and represents the weight coefficient;

[0064] Based on the matching cost, feature matching is performed on the mapping point and a number of adjacent feature points of the mapping point.

[0065] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0066] This invention utilizes a neural network model to adaptively learn complex spatial mapping relationships. Compared to traditional affine transformation methods, this mechanism can handle nonlinear transformations, effectively addressing the distortion problem caused by viewing angle variations in large-scale photovoltaic arrays. Furthermore, the invention combines a mapping point with a matching mechanism of adjacent feature points to select feature points that match the mapping point. Based on the feature point matching results, the Hungarian algorithm is used to further calibrate the matching relationship between the mapping point and its matching feature point, further improving the accuracy of photovoltaic device ID matching results. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0068] Figure 1 The figure is a flow chart of a photovoltaic device ID matching method based on anchor point affine according to an embodiment of the present invention.

[0069] Figure 2 This is a network structure diagram of a coordinate mapping model involved in an embodiment of the present invention.

[0070] Figure 3 Schematic diagram of the flow of the training method of the coordinate mapping model involved in an embodiment of the present invention.

[0071] Figure 4 The figure is a schematic structural diagram of a photovoltaic device ID matching system based on anchor point affine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] First, it should be noted that the terms used in the embodiments of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. As used in the embodiments of the present invention and the appended claims, the singular forms "a," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise.

[0073] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Therefore, it should be recognized by those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0074] It should be noted that in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than those described in this specification. In addition, a single step described in this specification may be broken down into multiple steps for description in other embodiments, and multiple steps described in this specification may be combined into a single step for description in other embodiments.

[0075] This specification aims to propose a solution to the problem of insufficient accuracy in existing photovoltaic device ID matching. Specifically, one or more embodiments of this specification propose a photovoltaic device ID matching method and system based on anchor point affine. This method uses a neural network model to adaptively learn complex spatial mapping relationships, effectively solving the distortion problem caused by changes in viewing angles in large-scale photovoltaic arrays. On this basis, this method combines the mapping point and adjacent feature point matching mechanism to screen feature points that match the mapping point, and based on the feature point matching results combined with the Hungarian algorithm, it further corrects the matching relationship between the mapping point and its matching feature point, further improving the accuracy of the photovoltaic device ID matching results.

[0076] The following will further describe in detail the satellite equipment status monitoring method and apparatus described in one or more embodiments of this specification in conjunction with the accompanying drawings and specific embodiments. However, this detailed description does not constitute a limitation on the embodiments of this specification.

[0077] Please refer to Figure 1 , Figure 1 A schematic flow chart of a photovoltaic device ID matching method based on anchor point affine is provided. The method includes steps S100 to S110:

[0078] S100: Acquire an aerial image of a photovoltaic array to be matched and a corresponding as-built image.

[0079] The photovoltaic array aerial image can be obtained by using an aerial drone to photograph the photovoltaic equipment installation site. The as-built image refers to a construction target map of the photovoltaic array installation site, with the as-built image annotated with information such as the installation location of each photovoltaic device in the target site and the device ID. The actual installation location of each photovoltaic device in the photovoltaic equipment installation site must comply with the photovoltaic device installation location standards in the as-built image.

[0080] S102: Extracting feature points of the photovoltaic equipment in the photovoltaic array aerial image and the photovoltaic equipment in the as-built image, and determining the coordinates of the feature points.

[0081] In some embodiments, the outline of the photovoltaic equipment in the photovoltaic array aerial image and the as-built image can be first detected by a target detection algorithm, and then one or more points that can characterize the position of the photovoltaic equipment can be extracted from the detection frame as feature points. Among them, a target detection algorithm based on Faster R-CNN, a target detection algorithm based on RetinaNet, or a target detection algorithm based on YOLO can be used to perform target detection on the photovoltaic equipment in the photovoltaic array aerial image and the as-built image. This embodiment does not limit the specific target detection algorithm. Similarly, this embodiment does not limit the feature point detection algorithm.

[0082] S104: Mapping the feature points in the as-built image to the photovoltaic array aerial image using a pre-trained coordinate mapping model to obtain coordinates of the mapping points.

[0083] Before implementing this step, the coordinate mapping model needs to be trained so that the coordinate mapping model has the ability to map feature points.

[0084] Please refer to Figure 2 In some embodiments, the coordinate mapping model can be used Figure 2 The network structure shown in Figure 2 is as follows. Figure 2 As shown, the coordinate mapping model includes a first model of a neural network model and a second model based on an interpolator.

[0085] Among them, the first model can adopt a multi-layer perceptron structure to learn the mapping relationship from source coordinates to target coordinates through nonlinear transformation. The second model can be implemented using an RBF interpolator, which is based on the thin plate spline (TPS) theory and achieves smooth interpolation by minimizing the surface bending energy. This method is particularly suitable for processing irregularly distributed data points and can ensure the accuracy at the interpolation point. In this embodiment, a hybrid model of a neural network model and an RBF (Radial Basis Function) interpolator is used for coordinate point mapping. The neural network model is used to ensure global transformation capability, and the RBF interpolator is used to ensure local accuracy, thereby ensuring the mapping accuracy of the coordinate mapping model.

[0086] Obviously, the structure of the coordinate mapping model used in this embodiment is not limited to Figure 2 The structure shown, other network models that can achieve the same function should also be within the protection scope of this embodiment.

[0087] Below, only Figure 2 Taking the coordinate mapping model shown in FIG. 1 as an example, the training process of the coordinate mapping model is explained.

[0088] Please refer to Figure 3 , Figure 3 The flowchart of the training method of the coordinate mapping model is schematically shown, and the training method includes S300 to S306.

[0089] S300: Acquire a photovoltaic array aerial sample image and an as-built sample image corresponding to the photovoltaic array aerial sample image.

[0090] It should be noted that this step aims to select multiple image pairs, each consisting of an as-built sample image and a corresponding PV array aerial sample image. Obviously, considering factors such as the shooting environment and shooting angle, the same as-built sample image can correspond to multiple PV array aerial sample images to form different image pairs, thereby expanding the sample richness.

[0091] S302: Select multiple feature points in the as-built sample image to form a feature point sample, and determine the coordinates of target feature points corresponding to the feature point sample in the photovoltaic array aerial sample image.

[0092] The input and output of the coordinate mapping model are both point coordinate data, so it is necessary to extract feature point samples from the as-built sample images and the photovoltaic array aerial photography sample images.

[0093] Specifically, multiple feature points can be selected from the as-built sample images of each image pair through feature point detection or manual annotation. The coordinates of each feature point are a feature point sample. Next, target feature points that match the aforementioned feature points are found from the PV array aerial sample images of the image pair. The positions of the target feature points can be manually annotated to obtain the position coordinates of the target feature points. Alternatively, feature points can be extracted from the PV array aerial sample images first, and then the feature points in the PV array aerial sample images are matched with the feature points selected from the as-built sample images. Based on the matching results, the target feature points that match the feature points are determined. The coordinate positions of these target feature points are the labels of the aforementioned feature point samples.

[0094] S304: Input the feature point samples into the coordinate mapping model to obtain predicted mapping point coordinates.

[0095] Please continue to refer to Figure 2 ,exist Figure 2 In the coordinate mapping model shown, the feature point samples are input into a multi-layer support vector machine and an RBF interpolator, respectively, to obtain a first coordinate mapping result output by the multi-layer support vector machine and a second coordinate mapping result output by the RBF interpolator. The first coordinate mapping result and the second coordinate mapping result are then weighted and combined to obtain the coordinates of the mapping point, namely:

[0096] ;

[0097] in, Represents the input feature point sample, that is, the coordinates (x, y) of the feature point p. represents the output of the multi-layer support vector machine, Represents the output of the RBF interpolator. and represents the weight coefficient, . represents the predicted coordinates of the mapping point, .

[0098] In this step, the RBF interpolator consists of the abscissa interpolator and the ordinate interpolator. The training process of the abscissa interpolator is:

[0099] First, a small number of source anchor points are selected from the as-built sample image (the source anchor points can be representative points, such as the center point of the photovoltaic device), and target anchor points that match the above anchor points are selected from the photovoltaic array aerial sample image to form a small number of anchor point pairs.

[0100] The horizontal coordinate of the source anchor point is used as the sampling point, and the horizontal coordinate of the target anchor point is used as the measurement value of the sampling point. That is, the interpolation function of the horizontal coordinate interpolator Needs to be satisfied ,in, represents the sampling point, express The measured value.

[0101] The basis function of the thin plate spline of the RBF interpolator is defined as:

[0102] ;

[0103] Where r represents the radial distance between two points in space.

[0104] Assuming the interpolation function is the superposition of a set of linear basis functions, then It can be expressed as:

[0105] ;

[0106] in, represents the weight coefficient, and K represents the number of anchor pairs.

[0107] Will It is expressed in matrix form as:

[0108] ;

[0109] Among them, A is a The interpolation matrix of A, each element in The value of is obtained according to the basis function, . W is the coefficient vector, , represents the measurement value vector, .

[0110] Based on the above sampling points and observation values, the coefficient vector W can be solved, and the final abscissa interpolation function expression is:

[0111] ;

[0112] Similarly, the expression of the vertical coordinate interpolation function can be obtained as:

[0113] ;

[0114] When the feature point sample is input into the RBF interpolator, the horizontal coordinate of the feature point sample is input into the horizontal coordinate interpolator, and the vertical coordinate of the feature point sample is input into the vertical coordinate interpolator, so that the horizontal coordinate prediction value and the vertical coordinate prediction value can be obtained. The horizontal coordinate prediction value and the vertical coordinate prediction value constitute the above-mentioned second coordinate mapping result.

[0115] S306: Constructing a loss function based on the predicted mapping point coordinates and the coordinates of the target feature point, and using the loss function to update the coordinate mapping model until the coordinate mapping model that meets the requirements is obtained.

[0116] When constructing the loss function, you can use the mean square error (MSE) as the loss function:

[0117] ;

[0118] Among them, N represents the number of feature point samples, Indicates the label of the feature point sample.

[0119] The above is the training process of the coordinate mapping model. After the coordinate mapping model is trained, the coordinate mapping model can be used to map the feature points in the as-built image to the photovoltaic array aerial image to obtain the coordinates of the mapping points.

[0120] S106: In the photovoltaic array aerial image, feature matching is performed between the mapping point and several adjacent feature points of the mapping point.

[0121] Because aerial images of photovoltaic arrays often suffer from complex factors such as uneven lighting and shadow interference, errors may exist between the mapped points obtained by the coordinate mapping model and the actual target points. To correct such errors, this embodiment proposes a matching evaluation mechanism using 3D feature fusion.

[0122] Specifically, for each mapping point, adjacent feature points around the mapping point are found in the aerial image of the photovoltaic array, and then the matching cost between the mapping point and these adjacent feature points is calculated. Based on the matching cost, adjacent feature points that may have a matching relationship with the mapping point are selected.

[0123] The matching cost is calculated as follows:

[0124] The Euclidean distance cost Distance_Cost between the mapping point and the adjacent feature point is calculated, where Distance_Cost represents the Euclidean distance between the mapping point and the adjacent feature point.

[0125] Calculate the relative position relationship cost Position_Cost between the mapping point and the adjacent feature point:

[0126] ;

[0127] Assume that the mapping point is , p represents the mapping point The original feature point corresponding to the as-built image is the mapping point obtained by mapping the feature point p in the as-built image to the photovoltaic array aerial image using the coordinate mapping model. nᵢ represents the i-th adjacent feature point around the original feature point p in the as-built image, i=1,2,…,k, k is a positive integer, d(p,nᵢ) represents the distance between the original feature point p and the adjacent feature point nᵢ; p' represents the adjacent feature point of the mapping point in the photovoltaic array aerial image, nᵢ' represents the i-th adjacent feature point around the adjacent feature point p', i=1,2,…,k, d'(p',nᵢ') represents the distance between the adjacent feature point p' and its adjacent feature point nᵢ';

[0128] Calculate the consistency cost Anchor_Cost of the anchor point between the mapping point and the adjacent feature point:

[0129] ;

[0130] in, represents the jth anchor point in the as-built image, Indicates the original feature point p and the anchor point in the completed image The distance ratio between them, j=1,2,…,m, m is a positive integer; Indicates that in the aerial image of the photovoltaic array The corresponding anchor point, Represents the adjacent feature point p' and the anchor point The distance ratio between

[0131] Calculate the matching cost:

[0132] ;

[0133] in, and represents the weight coefficient;

[0134] Based on the matching cost, feature matching is performed on the mapping point and a number of adjacent feature points of the mapping point.

[0135] The matching evaluation mechanism based on 3D feature fusion described above comprehensively considers three dimensions of features: Euclidean distance, relative positional relationship, and anchor point consistency during the matching process, constructing a reliable matching evaluation criterion through weighted fusion. Euclidean distance reflects the spatial proximity of point pairs, relative positional relationship ensures local structural consistency, and anchor point consistency provides a global constraint. This multi-dimensional feature fusion strategy effectively overcomes the susceptibility to interference associated with traditional single-feature matching, enabling the system to maintain stable matching results, particularly in complex scenarios such as uneven lighting and shadow interference.

[0136] S108: Taking the set of mapping points as a first point set, taking the feature points matching the mapping points as a second point set, and calculating an optimal matching result between the first point set and the second point set using the Hungarian algorithm.

[0137] After the feature matching in step S106, each mapping point may have a matching relationship with one or more adjacent feature points. For example, these matching relationships can be described as (A1, B1), (A1, B2), (A2, B2), (A2, B3), where A1 and A2 represent mapping points, (A1, B1) indicates that A1 may have a matching relationship with its adjacent feature point B1, (A1, B2) indicates that A1 may have a matching relationship with its adjacent feature point B2, (A2, B3) indicates that A2 may have a matching relationship with its adjacent feature point B3, and so on.

[0138] Based on this, in this step, the set of mapping points can be used as the first point set, and the feature points matching the mapping points can be used as the second point set. Then, the Hungarian algorithm is used to solve the optimal matching result between the first point set and the second point set.

[0139] S110: Based on the optimal matching result, the device ID corresponding to the mapping point is matched to the corresponding feature point in the photovoltaic array aerial image.

[0140] Before implementing this step, OCR recognition may be performed on the as-built image to extract the ID of the photovoltaic equipment in the as-built image.

[0141] After obtaining the optimal matching result, the feature point matched by the mapping point can be obtained, and then the device ID corresponding to the original feature point of the mapping point is matched to this feature point.

[0142] Corresponding to the above-mentioned photovoltaic device ID matching method based on anchor point affine, this embodiment further provides a photovoltaic device ID matching system based on anchor point affine, which is used to implement the above-mentioned photovoltaic device ID matching method based on anchor point affine. Figure 4 Said system comprises:

[0143] A data acquisition module is used to obtain the aerial image of the photovoltaic array to be matched and the corresponding as-built image;

[0144] a data preprocessing module, configured to extract feature points of the photovoltaic devices in the photovoltaic array aerial image and the photovoltaic devices in the as-built image, and determine the coordinates of the feature points;

[0145] A coordinate mapping module is used to map the feature points in the as-built image to the photovoltaic array aerial image using a pre-trained coordinate mapping model to obtain the coordinates of the mapping points;

[0146] The matching module is configured to perform feature matching between the mapping point and several adjacent feature points of the mapping point in the aerial image of the photovoltaic array; use the set of mapping points as a first point set, use the feature points matched with the mapping points as a second point set, and calculate an optimal matching result between the first point set and the second point set using the Hungarian algorithm; and match the device ID corresponding to the mapping point to the corresponding feature point in the aerial image of the photovoltaic array based on the optimal matching result.

[0147] Specifically, the aerial image of the photovoltaic array captured by the data acquisition module can be obtained by using an aerial drone to photograph the photovoltaic equipment installation site. The as-built image refers to a construction target map of the photovoltaic array installation at the photovoltaic equipment installation site. The as-built image is annotated with information such as the installation location of each photovoltaic device at the target site and the device ID. The actual installation location of each photovoltaic device at the photovoltaic equipment installation site must comply with the photovoltaic device installation location standards in the as-built image.

[0148] Specifically, the data preprocessing module can detect the outline of the photovoltaic equipment in the photovoltaic array aerial image and the as-built image through a target detection algorithm, and then extract one or more points in the detection frame that can characterize the position of the photovoltaic equipment as feature points. Among them, a target detection algorithm based on Faster R-CNN, a target detection algorithm based on RetinaNet, or a target detection algorithm based on YOLO can be used to perform target detection on the photovoltaic equipment in the photovoltaic array aerial image and the as-built image. This embodiment does not limit the specific target detection algorithm. Similarly, this embodiment does not limit the feature point detection algorithm.

[0149] Specifically, the coordinate mapping model includes a first model based on a neural network model and a second model based on an interpolator; the coordinate mapping module is specifically used to:

[0150] Inputting the feature points in the as-built image into the first model to obtain a first coordinate mapping result;

[0151] Inputting the feature points in the as-built image into the second model to obtain a second coordinate mapping result;

[0152] A weighted combination is performed on the first coordinate mapping result and the second coordinate mapping result to obtain the coordinates of the mapping point.

[0153] Specifically, the system further includes a training module, which is used to:

[0154] Acquire a photovoltaic array aerial sample image and an as-built sample image corresponding to the photovoltaic array aerial sample image;

[0155] Selecting multiple sets of feature point sets in the as-built sample image to form feature point samples, and determining coordinates of target feature points corresponding to the feature point samples in the photovoltaic array aerial sample image;

[0156] Inputting the feature point samples into the coordinate mapping model to obtain predicted mapping point coordinates;

[0157] A loss function is constructed based on the predicted mapping point coordinates and the coordinates of the target feature points, and the coordinate mapping model is updated using the loss function until the coordinate mapping model that meets the requirements is obtained.

[0158] Specifically, the data preprocessing module is further used to:

[0159] Perform OCR recognition on the as-built image to extract the ID of the photovoltaic equipment in the as-built image.

[0160] Specifically, the matching module is specifically used to:

[0161] Calculate the Euclidean distance cost Distance_Cost between the mapping point and the adjacent feature point;

[0162] Calculate the relative position relationship cost Position_Cost between the mapping point and the adjacent feature point:

[0163] ;

[0164] Wherein, p represents the original feature point corresponding to the mapping point in the as-built image, nᵢ represents the i-th adjacent feature point around the original feature point p in the as-built image, i=1,2,…,k, k is a positive integer, d(p,nᵢ) represents the distance between the original feature point p and the adjacent feature point nᵢ; p' represents the adjacent feature point of the mapping point in the PV array aerial image, nᵢ' represents the i-th adjacent feature point around the adjacent feature point p', i=1,2,…,k, d'(p',nᵢ') represents the distance between the adjacent feature point p' and its adjacent feature point nᵢ';

[0165] Calculate the consistency cost Anchor_Cost of the anchor point between the mapping point and the adjacent feature point:

[0166] ;

[0167] in, represents the jth anchor point in the as-built image, Indicates the original feature point p and the anchor point in the completed image The distance ratio between them, j=1,2,…,m, m is a positive integer; Indicates that in the aerial image of the photovoltaic array The corresponding anchor point, Represents the adjacent feature point p' and the anchor point The distance ratio between

[0168] Calculate the matching cost:

[0169] ;

[0170] in, and represents the weight coefficient;

[0171] Based on the matching cost, feature matching is performed on the mapping point and a number of adjacent feature points of the mapping point.

[0172] It should be understood that the structures illustrated in the embodiments of this specification do not constitute specific limitations on the systems of the embodiments of this specification. In other embodiments of the specification, the above-mentioned system may include more or fewer components than shown in the figure, or some components may be combined, some components may be separated, or the components may be arranged differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0173] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from other embodiments. In particular, the device embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0174] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0175] It should be noted that the above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many similar variations are possible. All variations directly derived from or associating with the present invention by those skilled in the art are intended to fall within the scope of protection of the present invention.

Claims

1. A photovoltaic equipment ID matching method based on anchor point affine, characterized in that: include: Obtaining the aerial image of the photovoltaic array to be matched and the corresponding as-built image; Extracting feature points of the photovoltaic equipment in the photovoltaic array aerial image and the photovoltaic equipment in the as-built image, and determining the coordinates of the feature points; Using a pre-trained coordinate mapping model, the feature points in the as-built image are mapped to the photovoltaic array aerial image to obtain the coordinates of the mapping points; In the photovoltaic array aerial image, performing feature matching between the mapping point and a plurality of adjacent feature points of the mapping point; Taking the set of mapping points as a first point set, taking the feature points matched with the mapping points as a second point set, and calculating the optimal matching result between the first point set and the second point set using the Hungarian algorithm; Based on the optimal matching result, the device ID corresponding to the mapping point is matched to the corresponding feature point in the photovoltaic array aerial image.

2. The method according to claim 1, characterized in that The coordinate mapping model includes a first model based on a neural network model and a second model based on an interpolator; Using a pre-trained coordinate mapping model, the feature points in the as-built image are mapped to the photovoltaic array aerial image to obtain the coordinates of the mapping points, specifically including: Inputting the feature points in the as-built image into the first model to obtain a first coordinate mapping result; Inputting the feature points in the as-built image into the second model to obtain a second coordinate mapping result; A weighted combination is performed on the first coordinate mapping result and the second coordinate mapping result to obtain the coordinates of the mapping point.

3. The method according to claim 2, characterized in that The method further comprises: Acquire a photovoltaic array aerial sample image and an as-built sample image corresponding to the photovoltaic array aerial sample image; Selecting a plurality of feature points in the as-built sample image, using the coordinates of the plurality of feature points as feature point samples, and determining the coordinates of target feature points corresponding to the feature points in the photovoltaic array aerial sample image; Inputting the feature point samples into the coordinate mapping model to obtain predicted mapping point coordinates; A loss function is constructed based on the predicted mapping point coordinates and the coordinates of the target feature points, and the coordinate mapping model is updated using the loss function until the coordinate mapping model that meets the requirements is obtained.

4. The method according to claim 1, wherein The method further comprises: Perform OCR recognition on the as-built image to extract the ID of the photovoltaic equipment in the as-built image.

5. The method according to claim 1, wherein In the photovoltaic array aerial image, feature matching is performed on the mapping point with several adjacent feature points of the mapping point, specifically including: Calculate the Euclidean distance cost Distance_Cost between the mapping point and the adjacent feature point; Calculate the relative position relationship cost Position_Cost between the mapping point and the adjacent feature point: ； Wherein, p represents the original feature point corresponding to the mapping point in the as-built image, nᵢ represents the i-th adjacent feature point around the original feature point p in the as-built image, i=1,2,…,k, k is a positive integer, d(p,nᵢ) represents the distance between the original feature point p and the adjacent feature point nᵢ; p' represents the adjacent feature point of the mapping point in the PV array aerial image, nᵢ' represents the i-th adjacent feature point around the adjacent feature point p', i=1,2,…,k, d'(p',nᵢ') represents the distance between the adjacent feature point p' and its adjacent feature point nᵢ'; Calculate the consistency cost Anchor_Cost of the anchor point between the mapping point and the adjacent feature point: ； in, represents the jth anchor point in the as-built image, Indicates the original feature point p and the anchor point in the completed image The distance ratio between them, j=1,2,…,m, m is a positive integer; Indicates that in the aerial image of the photovoltaic array The corresponding anchor point, Represents the adjacent feature point p' and the anchor point The distance ratio between Calculate the matching cost: ； in, and represents the weight coefficient; Based on the matching cost, feature matching is performed on the mapping point and a number of adjacent feature points of the mapping point.

6. A photovoltaic equipment ID matching system based on anchor point affine, characterized in that: include: A data acquisition module is used to obtain the aerial image of the photovoltaic array to be matched and the corresponding as-built image; a data preprocessing module, configured to extract feature points of the photovoltaic devices in the photovoltaic array aerial image and the photovoltaic devices in the as-built image, and determine the coordinates of the feature points; A coordinate mapping module is used to map the feature points in the as-built image to the photovoltaic array aerial image using a pre-trained coordinate mapping model to obtain the coordinates of the mapping points; The matching module is configured to perform feature matching between the mapping point and several adjacent feature points of the mapping point in the aerial image of the photovoltaic array; use the set of mapping points as a first point set, use the feature points matched with the mapping points as a second point set, and calculate an optimal matching result between the first point set and the second point set using the Hungarian algorithm; and match the device ID corresponding to the mapping point to the corresponding feature point in the aerial image of the photovoltaic array based on the optimal matching result.

7. The system according to claim 6, characterized in that The coordinate mapping model includes a first model based on a neural network model and a second model based on an interpolator; the coordinate mapping module is specifically used to: Inputting the feature points in the as-built image into the first model to obtain a first coordinate mapping result; Inputting the feature points in the as-built image into the second model to obtain a second coordinate mapping result; A weighted combination is performed on the first coordinate mapping result and the second coordinate mapping result to obtain the coordinates of the mapping point.

8. The system according to claim 7, characterized in that The system further comprises a training module, wherein the training module is configured to: Acquire a photovoltaic array aerial sample image and an as-built sample image corresponding to the photovoltaic array aerial sample image; Selecting multiple sets of feature point sets in the as-built sample image to form feature point samples, and determining coordinates of target feature points corresponding to the feature point samples in the photovoltaic array aerial sample image; Inputting the feature point samples into the coordinate mapping model to obtain predicted mapping point coordinates; A loss function is constructed based on the predicted mapping point coordinates and the coordinates of the target feature points, and the coordinate mapping model is updated using the loss function until the coordinate mapping model that meets the requirements is obtained.

9. The system according to claim 6, wherein: The data preprocessing module is also used for: Perform OCR recognition on the as-built image to extract the ID of the photovoltaic equipment in the as-built image.

10. The system according to claim 6, wherein: The matching module is specifically used for: Calculate the Euclidean distance cost Distance_Cost between the mapping point and the adjacent feature point; Calculate the relative position relationship cost Position_Cost between the mapping point and the adjacent feature point: ； Wherein, p represents the original feature point corresponding to the mapping point in the as-built image, nᵢ represents the i-th adjacent feature point around the original feature point p in the as-built image, i=1,2,…,k, k is a positive integer, d(p,nᵢ) represents the distance between the original feature point p and the adjacent feature point nᵢ; p' represents the adjacent feature point of the mapping point in the PV array aerial image, nᵢ' represents the i-th adjacent feature point around the adjacent feature point p', i=1,2,…,k, d'(p',nᵢ') represents the distance between the adjacent feature point p' and its adjacent feature point nᵢ'; Calculate the consistency cost Anchor_Cost of the anchor point between the mapping point and the adjacent feature point: ； in, represents the jth anchor point in the as-built image, Indicates the original feature point p and the anchor point in the completed image The distance ratio between them, j=1,2,…,m, m is a positive integer; Indicates that in the aerial image of the photovoltaic array The corresponding anchor point, Represents the adjacent feature point p' and the anchor point The distance ratio between Calculate the matching cost: ； in, and represents the weight coefficient; Based on the matching cost, feature matching is performed on the mapping point and a number of adjacent feature points of the mapping point.

Images