Photovoltaic project construction progress recognition method based on UAV images and neural network

Abstract

The inventor has proposed a method for identifying the construction progress of a photovoltaic project based on drone images and neural networks. The method first divides the photovoltaic project plant into several square arrays, and uses the square array as a unit to shoot construction progress site images based on drones according to preset route tasks; for the site images of each square array, the orthophoto images of the corresponding square array are obtained based on the relevant orthophoto image generation algorithm, and the obtained images are cropped and format converted to obtain suitable photovoltaic project construction progress pictures, and the project progress recognition model obtained by training based on the optimized YOLT algorithm model is recognized, and the recognition number and design number of the square array pile foundation, photovoltaic bracket, and photovoltaic module are compared, and finally the actual project construction progress is obtained. The method provided by the present invention greatly improves the efficiency of the construction progress statistics of photovoltaic projects, and is of great significance for project managers to grasp the project construction progress and formulate subsequent construction plans.

Description

Technical Field

[0001] The present invention relates to a photovoltaic project construction progress recognition method based on drone images and neural networks Background Art

[0002] Large-scale photovoltaic power stations are mostly built in remote areas. This poses a great challenge to project managers in grasping the construction progress and understanding the on-site conditions. At present, the management of construction progress mostly adopts the method of collecting relevant data on-site manually. However, this method is time-consuming and consumes a lot of human resources. In addition, due to the large area of ​​the photovoltaic power station and the large number of photovoltaic modules, this method is often accompanied by human error, which has a great adverse impact on project managers in grasping the construction progress. Summary of the invention

[0003] The purpose of the present invention is to provide a photovoltaic project construction progress identification method based on drone images and neural networks to improve the efficiency of photovoltaic project construction progress statistics and solve the problems of long time consumption and high human resource consumption in the above-mentioned manual statistics of photovoltaic project construction progress. To this end, the present invention adopts the following technical solutions:

[0004] The photovoltaic project construction progress recognition method based on UAV images and neural networks includes the following steps:

[0005] Step 1: Divide the photovoltaic project plant area into several square arrays, and use pre-deployed drones to take timed photos of the construction progress of the square arrays through preset flight missions to obtain image collection data;

[0006] Step 2: Generate an orthophoto of the project progress of the corresponding square array based on the orthophoto generation algorithm using the acquired image acquisition data;

[0007] Step 3: Project the project progress orthophoto image to the geographic coordinate system corresponding to the photovoltaic project design drawing, crop the project progress orthophoto image according to the coordinate range of the square design drawing, convert the cropped project progress orthophoto image into a picture format, and obtain a construction progress picture;

[0008] Step 4: Use multiple construction progress pictures as training sets to train the optimized YOLT algorithm model to obtain a photovoltaic project construction progress recognition model, wherein the multiple construction progress pictures cover all construction completion states to be identified, and the construction completion states to be identified include pile foundation construction completion, bracket construction completion, and photovoltaic module construction completion. Based on the photovoltaic project construction progress recognition model, the construction progress pictures of all square arrays in the photovoltaic project plant are identified and processed to obtain the recognition results of the construction completion state of each square array, including the identification number of pile foundation construction completion, bracket construction completion, and photovoltaic module construction completion; the optimized YOLT algorithm model is an algorithm model formed by using the YOLT algorithm model as the basic framework and rewriting the non-maximum suppression module in the basic framework;

[0009] Step 5: Save the identification results of the construction completion status of each array to the database; call the identification results stored in the database, compare them with the design quantities of pile foundation, support construction and photovoltaic components, and obtain the construction progress of the entire photovoltaic project.

[0010] On the basis of adopting the above technical solutions, the present invention may also adopt the following further technical solutions, or use these further technical solutions in combination:

[0011] In step one, the image acquisition data obtained by the drone is stored in a cloud server, and the cloud server configures a post-notification of an access interface; the post-notification is monitored, and the real-time image acquisition data of the photovoltaic project construction site is downloaded using the access interface.

[0012] In step 2, the parameters of the orthophoto generation algorithm for generating orthophotos are adjusted to adapt to the ground texture characteristics and generate a project progress orthophoto that meets the identification requirements of the YOLT algorithm model.

[0013] In step three, it specifically includes: using reprojection to project the project progress orthophoto image to the geographic coordinate system corresponding to the photovoltaic project design drawing, to obtain an orthophoto image with the same geographic coordinate system code; for the orthophoto image containing the geographic coordinate system code, cropping the project progress orthophoto image according to the coordinate range of the square design drawing, converting the cropped orthophoto image into a picture format, to obtain a construction progress picture.

[0014] The optimized YOLT algorithm model is based on the YOLT algorithm model as the basic framework and the actual construction sequence logic as the criterion. The non-maximum suppression module in the YOLT algorithm framework is rewritten to be suitable for the photovoltaic project construction progress identification model.

[0015] In step 4, for the optimized YOLT algorithm model, by adjusting the training parameters and training with construction progress pictures, a photovoltaic project construction progress recognition model is obtained; the training parameters include image size, training rounds, and batch size.

[0016] Based on the actual construction sequence logic, the non-maximum suppression module in the YOLT algorithm framework is rewritten, including:

[0017] For the actual construction progress identification of photovoltaic projects, the priority is set as photovoltaic module construction>support construction>pile foundation construction; based on the priority order, the non-maximum suppression module in the YOLT algorithm framework is rewritten; the identification logic of the construction progress of the non-maximum suppression module is: starting to build the support indicates that the pile foundation construction has been completed, and starting to lay photovoltaic modules indicates that both the pile foundation construction and the support construction have been completed. Within the pixel range corresponding to the same photovoltaic string, only the boundary boxes of the identified high-priority construction projects that have been completed are retained. The photovoltaic string includes a photovoltaic module, a photovoltaic support and its pile foundation.

[0018] On the basis of retaining the completed bounding boxes of high-priority construction projects, the filtering rule of the non-maximum suppression module is as follows: for the repeatedly identified PV strings, filtering is performed according to the area of ​​the bounding box as the criterion, that is, for the identification result bounding boxes of the same PV string, only the bounding box with the largest area is retained to achieve the purpose of complete identification of the PV string.

[0019] In step five, the corresponding front-end service is developed based on the optimized YOLT algorithm model, and the identification results of the completion status of each square array are saved to the database; the identification results stored in the database are called through the front-end call interface, and compared with the design quantity of pile foundation, bracket construction and photovoltaic components to obtain the construction progress of the entire photovoltaic project. In order to solve the defects of the method of manually counting the construction progress of photovoltaic projects, the inventor proposes a photovoltaic project construction progress identification method based on drone images and neural networks. The method first divides the photovoltaic project plant into several square arrays, and takes the construction progress site images based on the square array as a unit based on the drone according to the preset route task; for the site images of each square array, the orthophoto image of the corresponding square array is obtained based on the relevant orthophoto image generation algorithm, and the obtained image is cropped and format converted to obtain a suitable photovoltaic project construction progress picture, and the project progress identification model obtained by training based on the optimized YOLT algorithm model is identified, and the identification quantity and design quantity of the square array pile foundation, photovoltaic bracket, and photovoltaic component are compared, and finally the actual project construction progress is obtained. The method provided by the present invention greatly improves the efficiency of photovoltaic project construction progress statistics, which is of great significance for project managers to grasp the project construction progress and formulate subsequent construction plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a photovoltaic project construction progress identification method based on drone images and neural networks provided in an embodiment of the present invention.

[0021] Figure 2 A logical schematic diagram of the recognition of the construction completion status by the non-maximum suppression module provided in an embodiment of the present invention.

[0022] Figure 3 A schematic diagram of the process of identifying a progress picture using the YOLT algorithm model provided in an embodiment of the present invention.

[0023] Figure 4 A schematic diagram comparing the photovoltaic string recognition results of the YOLT algorithm model before and after optimization provided in an embodiment of the present invention when the bracket is in a completed state.

[0024] Figure 5 A schematic diagram of a preferred photovoltaic project progress identification result provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method and technical scheme of a photovoltaic project construction progress identification method based on drone images and neural networks proposed by the present invention in combination with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Unless otherwise defined, all technical and scientific terms used in this article have the same meaning as those generally understood by technicians in the technical field of the present invention.

[0026] Reference Figure 1-2 The present invention provides a photovoltaic project construction progress identification method based on drone images and neural networks, comprising the following steps:

[0027] Step 1: Divide the photovoltaic project plant area into several square arrays, and based on the pre-deployed drones, take regular photos of the construction progress of the square arrays through preset flight missions to obtain image acquisition data; further, the image acquisition data obtained by the drone is stored in the cloud server, and the cloud server configures the post-notification of the access interface; monitor the post-notification, and use the access interface to download the real-time image acquisition data of the photovoltaic project construction site. The post-notification is a unified resource downloader of photovoltaic project construction progress images provided by the cloud server to realize the transmission of the captured images.

[0028] Step 2: Generate an orthophoto of the project progress of the corresponding square matrix based on the orthophoto generation algorithm using the acquired image acquisition data. Furthermore, in this embodiment, the parameters of the orthophoto generation algorithm for generating the orthophoto are adjusted, including but not limited to the position of the drone camera, the pitch angle and other parameters, and the photos are fixed in a fixed position to adapt to the ground texture characteristics, so as to construct a high-quality orthophoto of the project progress that meets the identification requirements of the YOLT algorithm model.

[0029] Step three: project the project progress orthophoto image to the geographic coordinate system corresponding to the photovoltaic project design drawing, crop the project progress orthophoto image according to the coordinate range of the square design drawing, and convert the cropped project progress orthophoto image into a conventional image format, including but not limited to JPG, PNG, to obtain a construction progress image.

[0030] Exemplarily, the geographic coordinate system corresponding to the photovoltaic project design drawing can be the CGCS2000 coordinate system, i.e., the 2000 National Geodetic Coordinate System. The present invention adopts a reprojection method to project the orthophoto image to obtain an orthophoto image with the same geographic coordinate system code as the design drawing, i.e., the EPSG code; in order to cover the complete array in the shooting range of the drone, after the shooting of the designated array is completed, a certain area will be surrounded by shooting outside the area. The orthophoto image needs to be cropped according to the coordinate range of the array's design drawing to obtain an orthophoto image with the same range as the design drawing.

[0031] In this embodiment, step three specifically includes:

[0032] (1) Projecting the project progress orthophoto image to the geographic coordinate system corresponding to the photovoltaic project design drawing by reprojection to obtain an orthophoto image with the same geographic coordinate system encoding;

[0033] (2) For orthophoto images containing geographic coordinate system codes, the project progress orthophoto images are cropped according to the coordinate range of the square design drawing. After cropping, an orthophoto image with the same range as the design drawing is obtained, which is converted into a picture format to obtain a construction progress picture.

[0034] Step 4: Use multiple construction progress pictures as training sets to train the optimized YOLT algorithm model to obtain a photovoltaic project construction progress recognition model, wherein the multiple construction progress pictures cover all construction completion states to be identified, and the construction completion states to be identified include pile foundation construction completion, bracket construction completion, and photovoltaic module construction completion. Based on the photovoltaic project construction progress recognition model, the construction progress pictures of all square arrays in the photovoltaic project plant are identified and processed to obtain the recognition results of the construction completion state of each square array, including the identification number of pile foundation construction completion, bracket construction completion, and photovoltaic module construction completion; the optimized YOLT algorithm model is an algorithm model formed by using the YOLT algorithm model as the basic framework and rewriting the non-maximum suppression module in the basic framework;

[0035] In the present invention, covering all the construction completion states to be identified in the multiple construction progress pictures means that the construction progress pictures in the training set contain images of all the construction completion states to be identified. For example, some pictures already contain images of these three construction completion states, and some only include images of one or two construction completion states.

[0036] The optimized YOLT algorithm model takes the YOLT algorithm model as the basic framework and the actual construction sequence logic of the photovoltaic project as the criterion. The non-maximum suppression module in the YOLT algorithm framework is rewritten to be suitable for the photovoltaic project construction progress recognition model, which can better solve the photovoltaic string recognition problem related to the photovoltaic project construction in the remote sensing image.

[0037] The non-maximum suppression algorithm is an algorithm technology in image processing. Its main function is to remove the redundant bounding boxes identified, retain only the bounding boxes that are most likely to contain photovoltaic strings, and retain the best recognition results. In the present invention, the non-maximum suppression module in the YOLT algorithm framework is rewritten based on the actual construction sequence logic to achieve the invention purpose of accurately identifying the construction progress. Specifically including:

[0038] (1) Figure 3 The construction progress picture of the square shown in a in the figure is cut into a specified size by windowing. Figure 3 The project progress recognition model recognizes the sub-pictures shown in b1 and b2, and finally converts them into Figure 3 The recognition results of the sub-images b1 and b2 are spliced ​​to obtain Figure 3 The recognition result of the entire square matrix is ​​shown in c.

[0039] (2) The algorithm's recognition logic for the construction completion status is as follows: within the pixel range corresponding to the same PV string, if a PV module is identified, the identification results of the PV bracket and pile foundation within the pixel range are filtered out, and it is assumed that the PV bracket and pile foundation of the PV string are completed; for the part within the pixel range where the PV module is not identified, it is determined whether the PV bracket is identified. If the PV bracket is identified, the pile foundation identification result is filtered out, and it is assumed that the pile foundation of the PV string is completed; for the part within the pixel range where the PV module and PV bracket are not identified, the pile foundation is identified and the identification result is output.

[0040] (3) For sub-image stitching, there may be multiple recognition results within the pixel range corresponding to the same photovoltaic string, such as Figure 3 Taking the sub-images b1 and b2 as an example, two results b3 and b4 are identified for the overlapping parts of sub-images b1 and b2. When stitching, compare the bounding box areas of the same PV string and only keep the Figure 3 The bounding box with the largest area shown in c1 is used as the final recognition result of the PV string.

[0041] For the actual construction progress identification of photovoltaic projects, the priority is set as photovoltaic module construction>support construction>pile foundation construction; based on the priority order, the non-maximum suppression module in the YOLT algorithm framework is rewritten; the identification logic of the construction progress of the non-maximum suppression module is: starting to build the support means that the pile foundation construction has been completed, and starting to lay photovoltaic modules means that both the pile foundation construction and the support construction have been completed. Within the pixel range corresponding to the same photovoltaic string, only the boundary boxes of the identified high-priority construction projects that have been completed are retained.

[0042] On the basis of retaining the completed construction projects with high priority, the filtering rule of the non-maximum suppression module is as follows: for the repeatedly identified PV strings, they are filtered according to the area of ​​the bounding box as the criterion, that is, for the identification result bounding box of the same PV string, only the bounding box with the largest area is retained to achieve the purpose of completely identifying the PV string.

[0043] For the optimized YOLT algorithm model, by adjusting the training parameters and training with construction progress pictures, a photovoltaic project construction progress recognition model is obtained; the training parameters include image size, training rounds, and batch size. The photovoltaic project construction progress recognition model is packaged into an image and deployed to the GPU server for operation. The drone-photographed images of the corresponding array are downloaded by monitoring the post-notification. The images are processed into construction progress pictures through the above-mentioned image processing, and input into the photovoltaic project construction progress recognition model for recognition, and the recognition results of the construction completion status of each array are obtained, including the recognition number of poured pile foundations in the array, the recognition number of photovoltaic brackets erected in the array, and the recognition number of photovoltaic modules laid in the array.

[0044] Step 5: Save the identification results of the construction completion status of each square array to the database; call the identification results stored in the database and compare them with the design quantity of pile foundation, support construction and photovoltaic components to obtain the construction progress of the entire photovoltaic project. The details are as follows:

[0045] (1) The identification results of all arrays within the photovoltaic project plant are stored in a database, wherein the identification results include the identification results of the construction completion status of each array, including the identified number of pile foundations that have been cast in the array, the identified number of photovoltaic brackets that have been erected in the array, and the identified number of photovoltaic modules that have been laid in the array.

[0046] (2) The front end calls the identification results of all the arrays in the photovoltaic project plant area and the designed number of photovoltaic modules, photovoltaic brackets and pile foundations in each array stored in the database to obtain the quantitative indicators of the construction progress of the photovoltaic project. The quantitative indicators of the construction progress of the photovoltaic project include the ratio of the number of completed pile foundations to the designed number of pile foundations, the ratio of the number of completed photovoltaic brackets to the designed number of photovoltaic brackets, and the ratio of the number of completed photovoltaic modules to the designed number of photovoltaic modules.

[0047] Furthermore, in this embodiment, a corresponding front-end service is developed based on the optimized YOLT algorithm model, and the construction progress of the entire photovoltaic project is obtained through the front-end calling interface, including orthophotos of the array construction progress, the ratio of the number of poured pile foundations in the array to the designed number of pile foundations, the ratio of the number of photovoltaic brackets erected in the array to the designed number of photovoltaic brackets, and the ratio of the number of photovoltaic modules laid in the array to the designed number of photovoltaic modules.

[0048] During implementation, when starting the pre-deployed drones, clear weather with good light and low wind speed should be selected for shooting; after starting the drones stored at the airport, they should be equipped with cameras to fly along the established shooting strategies and shooting routes in the divided squares. When reaching the designated position, the lens shooting is set to shoot vertically downward. The shooting content includes but is not limited to the roads, pile foundations, photovoltaic brackets, and photovoltaic modules within the photovoltaic plant; after the drones have completed shooting the photovoltaic project, they will store the data in the cloud server and configure the access interface to obtain the project progress images.

[0049] The non-maximum suppression module reconstruction of the present invention is further described in detail below with reference to specific examples.

[0050] The biggest difficulty in identifying the progress of PV projects is that drone images are large in size, while PV strings are small, especially pile foundations, which tend to be clustered together from drone images and are difficult to identify.

[0051] The solution is to cut the original image into pieces, then input them into the model for recognition and fuse the recognition results of different scales. The recognition process is as follows: Figure 2and Figure 3 shown. Figure 3 In the figure a, the original drone image is cropped to a sub-image of a specified size because the image resolution is too large. Figure 3 b1 and Figure 3 b2 in ) as the input of the model. Adjacent cropped sub-images are guaranteed to have 40% overlap (e.g. Figure 3 The purpose is to ensure that every photovoltaic module, photovoltaic support and pile foundation in the photovoltaic project plant can be fully identified. This will lead to some repeated identification, such as Figure 3 b3 and Figure 3 As shown in b4 in the figure, the 40% overlapping parts after the combination were identified twice. The traditional non-maximum suppression algorithm filters out the PV strings that are repeatedly identified by comparing the confidence. However, in order to retain the complete identification results of the components as much as possible, this method is not ideal for PV projects. Therefore, the filtering rules of the non-maximum suppression algorithm are reconstructed, and the PV strings that are repeatedly identified are filtered according to the area of ​​the bounding box as the criterion. That is, within the pixel range corresponding to the same PV string (such as Figure 3 The bounding box with the largest area is retained to achieve the purpose of completely identifying the PV strings.

[0052] Secondly, due to the particularity of photovoltaic string identification from the perspective of drones, the upper components are likely to block the lower components from the perspective during aerial photography. Figure 3 As shown in a and b, for the part where the photovoltaic panels have been laid, the photovoltaic brackets and pile foundations below are completely invisible in the drone image. The occlusion of the pile foundation by the completed photovoltaic brackets and the photovoltaic brackets under construction is also quite different, as shown in Figure 1. Figure 4 As shown in d.

[0053] The non-maximum suppression algorithm used by the original YOLT sorts the confidence of the bounding boxes. The bounding boxes with confidence greater than a certain threshold are deleted. However, in the photovoltaic string recognition task from the perspective of the drone, it is difficult to effectively identify the obscured lower structure using this algorithm. Especially for the similar Figure 4 In the drone image under complex working conditions shown in d, the supports in some areas have been completed, while the supports in some areas have not yet been completed. It is very difficult to accurately identify the number of pile foundations under occlusion. Therefore, the non-maximum suppression algorithm is improved. The confidence of the bounding box is no longer used as the filtering rule. Instead, the non-maximum suppression algorithm is reconstructed with "class" as the judgment logic. That is, in the construction logic, starting to build the support means that the pile foundation has been completed, and starting to lay photovoltaic panels means that both the pile foundation and the support have been completed. For example, for the reconstructed non-maximum suppression algorithm, the photovoltaic support class and the pile foundation class are simultaneously identified within the pixel range corresponding to the photovoltaic string, such as Figure 4The bounding box of e in , that is, filtering out the bounding box of the pile foundation class identified by the photovoltaic string (such as Figure 4 f), in the background progress statistics, it is assumed that the pile foundation at this location is completed. The logic of photovoltaic modules is the same.

[0054] Based on the above two bounding box screening criteria, the non-maximum suppression module in the YOLT model is reconstructed. Figure 5 The project progress identification results of a photovoltaic project array using the method of the present invention are given. The results include the ratio of the number of pile foundations identified to the number of designed pile foundations is 2192 / 2192, and the pile foundation has been completed; the ratio of the number of photovoltaic brackets identified to the number of designed photovoltaic brackets is 197 / 274, and the photovoltaic brackets have been completed by 71.9%; the ratio of the number of photovoltaic modules identified to the number of designed photovoltaic brackets is 183 / 274, and the photovoltaic brackets have been completed by 66.8%. The accuracy of the quantitative results of each construction step meets the engineering requirements.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A photovoltaic project construction progress recognition method based on drone images and neural networks, characterized in that: The following steps are involved: Step 1: Divide the photovoltaic project plant area into several square arrays, and use pre-deployed drones to take timed photos of the construction progress of the square arrays through preset flight missions to obtain image collection data; Step 2: Generate an orthophoto of the project progress of the corresponding square array based on the orthophoto generation algorithm using the acquired image acquisition data; Step 3: Project the project progress orthophoto image to the geographic coordinate system corresponding to the photovoltaic project design drawing, crop the project progress orthophoto image according to the coordinate range of the square design drawing, convert the cropped project progress orthophoto image into a picture format, and obtain a construction progress picture; Step 4: Use multiple construction progress pictures as training sets to train the optimized YOLT algorithm model to obtain a photovoltaic project construction progress recognition model, wherein the multiple construction progress pictures cover all construction completion states to be identified, and the construction completion states to be identified include pile foundation construction completion, bracket construction completion, and photovoltaic module construction completion. Based on the photovoltaic project construction progress recognition model, the construction progress pictures of all square arrays in the photovoltaic project plant are identified and processed to obtain the recognition results of the construction completion state of each square array, including the identification number of pile foundation construction completion, bracket construction completion, and photovoltaic module construction completion; the optimized YOLT algorithm model is an algorithm model formed by using the YOLT algorithm model as the basic framework and rewriting the non-maximum suppression module in the basic framework; Step 5: Save the identification results of the construction completion status of each array to the database; call the identification results stored in the database, compare them with the design quantities of pile foundation, support construction and photovoltaic components, and obtain the construction progress of the entire photovoltaic project.

2. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 1 is characterized in that: In step one, the image acquisition data obtained by the drone is stored in a cloud server, and the cloud server configures a post-notification of an access interface; the post-notification is monitored, and the real-time image acquisition data of the photovoltaic project construction site is downloaded using the access interface.

3. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 1 is characterized in that: In step 2, the parameters of the orthophoto generation algorithm for generating orthophotos are adjusted to adapt to the ground texture characteristics and generate a project progress orthophoto that meets the identification requirements of the YOLT algorithm model.

4. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 1 is characterized in that: In step three, it specifically includes: Reprojecting the project progress orthophoto image to the geographic coordinate system corresponding to the photovoltaic project design drawing to obtain an orthophoto image with the same geographic coordinate system encoding; For the orthophoto images containing geographic coordinate system codes, the project progress orthophoto images are cropped according to the coordinate range of the square design drawing, and the cropped orthophoto images are converted into picture format to obtain the construction progress picture.

5. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 1 is characterized in that: The optimized YOLT algorithm model is based on the YOLT algorithm model as the basic framework and the actual construction sequence logic as the criterion. The non-maximum suppression module in the YOLT algorithm framework is rewritten to be suitable for the photovoltaic project construction progress identification model.

6. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 1 is characterized in that: In step 4, for the optimized YOLT algorithm model, by adjusting the training parameters and training with construction progress pictures, a photovoltaic project construction progress recognition model is obtained; the training parameters include image size, training rounds, and batch size.

7. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 5 is characterized in that: Based on the actual construction sequence logic, the non-maximum suppression module in the YOLT algorithm framework is rewritten, including: For the actual construction progress identification of photovoltaic projects, the priority is set as photovoltaic module construction>support construction>pile foundation construction; based on the priority order, the non-maximum suppression module in the YOLT algorithm framework is rewritten; the identification logic of the construction progress of the non-maximum suppression module is: starting to build the support indicates that the pile foundation construction has been completed, and starting to lay photovoltaic modules indicates that both the pile foundation construction and the support construction have been completed. Within the pixel range corresponding to the same photovoltaic string, only the boundary boxes of the identified high-priority construction projects that have been completed are retained. The photovoltaic string includes a photovoltaic module, a photovoltaic support and its pile foundation.

8. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 7 is characterized in that: On the basis of retaining the completed bounding boxes of high-priority construction projects, the filtering rule of the non-maximum suppression module is as follows: for the repeatedly identified PV strings, filtering is performed according to the area of ​​the bounding box as the criterion, that is, for the identification result bounding boxes of the same PV string, only the bounding box with the largest area is retained to achieve the purpose of complete identification of the PV string.

9. The photovoltaic project construction progress identification method based on drone images and neural networks according to claim 1 is characterized in that: In step five, a corresponding front-end service is developed based on the optimized YOLT algorithm model, and the identification results of the construction completion status of each array are saved in the database; through the front-end calling interface, the identification results stored in the database are called and compared with the design quantity of pile foundation, bracket construction and photovoltaic components to obtain the construction progress of the entire photovoltaic project.

Images