A method for identifying salt precipitation areas of photovoltaic panels in offshore floating photovoltaic power stations

Abstract

This invention relates to the field of photovoltaic panel image processing technology, specifically to a method for identifying salt precipitation areas in offshore floating photovoltaic power plants. The method includes the following steps: using a drone to capture an overall image of the offshore floating photovoltaic power plant, obtaining images of the power plant from drone inspections; performing photovoltaic panel region segmentation on the photovoltaic power plant images to identify and locate the position information of each photovoltaic panel, and cropping the original image based on this position information to obtain images of all individual photovoltaic panels; processing the individual photovoltaic panel images using a salt precipitation edge marking algorithm to obtain coarse salt precipitation edge marking maps for each panel; performing salt precipitation area identification and marking to obtain salt precipitation segmentation marking maps; and stitching all salt precipitation segmentation marking maps together to obtain an overall salt precipitation marking map of the photovoltaic power plant. This invention reduces light scattering and power generation efficiency degradation caused by salt precipitation, significantly improving the maintenance efficiency of offshore floating photovoltaic power plants for maintenance teams.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic panel image processing technology, and in particular to a method for identifying salt precipitation areas on photovoltaic panels in a floating offshore photovoltaic power station. Background Technology

[0002] Offshore floating photovoltaic (PV) power plants face different challenges than onshore PV power plants due to their unique geographical location and installation environment, especially the problem of salt deposition on the photovoltaic panels. Salt deposition is a common fouling phenomenon in offshore floating power plants, formed by the accumulation of salt on the surface of the photovoltaic panels after seawater evaporates. This widespread light-transmitting fouling not only hinders the normal convergence of light but also causes light scattering, thus significantly reducing the power generation efficiency of the PV power plant. Due to the complexity of the marine environment, such as wind, waves, humidity, and salt spray, these factors further exacerbate the formation and accumulation of salt deposition, posing a severe challenge to the maintenance and power generation efficiency of the PV power plant.

[0003] While drone technology offers a viable solution for the inspection of offshore floating photovoltaic power plants, the extreme difficulty of manual inspections means that maintenance work relies heavily on drones, which limits the efficiency and effectiveness of these inspections. Furthermore, due to the differences in color and distribution characteristics between salt deposits and common dust, traditional dust identification algorithms are not applicable to salt deposit problems, making the identification and treatment of salt deposit damage even more difficult. Currently, salt deposit areas on photovoltaic panels still require manual labeling and recording, which not only increases the difficulty of maintenance but also significantly increases workload and cost. Therefore, existing technologies have significant shortcomings in efficiently and accurately identifying and addressing salt deposit problems in offshore floating photovoltaic power plants. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides a method for identifying salt precipitation areas of photovoltaic panels in a floating photovoltaic power station at sea.

[0005] A method for identifying salt deposition areas of photovoltaic panels in a floating offshore photovoltaic power station includes the following steps:

[0006] S1: Use drones to take overall pictures of the floating photovoltaic power station at sea, and obtain images of the photovoltaic power station obtained by drone inspection;

[0007] S2: Perform photovoltaic panel region segmentation on the photovoltaic power station image using a target detection algorithm to identify and locate the position information of each photovoltaic panel, and crop the original image based on the position information to obtain images of all individual photovoltaic panels;

[0008] S3: The salt precipitation edge marking algorithm is used to process the image of a single photovoltaic panel to obtain a coarse salt precipitation edge marking map of the single panel;

[0009] S4: Combine all the coarse marking maps of salt precipitation edges with the original image to identify and mark the salt precipitation region, and obtain the salt precipitation segmentation marking map;

[0010] S5: Stitch together all the salt precipitation segmentation marker images to obtain the overall salt precipitation marker image of the photovoltaic power station.

[0011] Furthermore, the target detection algorithm in S2 uses a Faster R-CNN neural network. The photovoltaic power station image is imported into the trained Faster R-CNN neural network to obtain the location region of the photovoltaic panel.

[0012] Furthermore, the Faster R-CNN neural network specifically includes:

[0013] Data preprocessing: The overall photovoltaic power station images captured by the drone are preprocessed, including adjusting the image size to meet the input requirements of the Faster Rcnn network and performing color normalization to reduce the impact of changes in illumination.

[0014] Image input: The preprocessed image is input into the trained Faster R-CNN neural network, which combines a region proposal network and a deep learning model of R-CNN to detect the location and category of the target object from the image;

[0015] Region Proposal: In Faster RCN, a series of region proposals are generated through RPN. RPN evaluates each location in the image and proposes candidate bounding boxes for regions containing photovoltaic panels;

[0016] Feature extraction: For each region proposal, its features are extracted through a shared convolutional layer. The convolutional layer extracts rich feature representations from the image to identify and distinguish photovoltaic panels.

[0017] Classification and Regression: The extracted features are fed into subsequent network layers for classification and bounding box regression;

[0018] Non-maximum suppression: The non-maximum suppression algorithm is applied to eliminate proposal boxes with excessive overlap and retain the best proposal box as the final photovoltaic panel position;

[0019] Output: The Faster R-CNN neural network outputs the location information of each photovoltaic panel, including the coordinates of the bounding box. This location information is used to crop the original image to obtain individual photovoltaic panel images for subsequent analysis and processing.

[0020] Furthermore, S3 specifically includes:

[0021] S31: To perform noise reduction processing on the photovoltaic panel image;

[0022] S32: Perform HLS color space conversion on the denoised image of the photovoltaic panel;

[0023] S33: Based on the characteristics of the salt precipitation region in the HLS color space, perform region marking processing to obtain the salt precipitation region map;

[0024] S34: In the obtained salt precipitation region map, fill the salt precipitation region with white, perform edge recognition processing on the filled region, and obtain a coarse salt precipitation edge map.

[0025] Furthermore, the noise reduction process in S31 includes performing Gaussian blur processing on the image, as follows:

[0026]

[0027] Where x is the horizontal coordinate of the image, y is the vertical coordinate of the image, P(x,y) is the pixel value at coordinate (x,y) in the processed image, I(x,y) is the pixel value at coordinate (x,y) in the original image, and σ is the preset variance value used to adjust the degree of Gaussian blur.

[0028] Furthermore, the HLS color space conversion in S32 is specifically represented as follows:

[0029]

[0030] Where R, G, and B represent the pixel values ​​of the red, green, and blue channels in the RGB color space, respectively; H, L, and S represent the hue, brightness, and saturation values ​​in the HLS color space, respectively; and max and min represent the maximum and minimum values ​​of the three channels in the RGB color space, respectively.

[0031] Furthermore, the region marking process in S33 is as follows:

[0032]

[0033] Where New(x,y) is the pixel value at (x,y) of the processed image, I(x,y) is the pixel value at (x,y) of the denoised image, H(x,y) is the pixel value at (x,y) in the H channel, L(x,y) is the pixel value at (x,y) in the L channel, S(x,y) is the pixel value at (x,y) in the S channel, and H... mean The mean of the H channel, L mean S is the mean of the L channel. mean This represents the mean of the S-channel.

[0034] Furthermore, S4 includes combining the original image and all salt precipitation edge coarse marking images, and using a standard FCN fully convolutional neural network to identify and mark the salt precipitation region. The input of the FCN fully convolutional neural network is changed from the original three-channel RGB color space to a four-channel input, and the fourth channel is the grayscale data of the salt precipitation edge coarse marking image, so as to obtain the salt precipitation segmentation marking image of each photovoltaic panel.

[0035] The beneficial effects of this invention are:

[0036] This invention employs drone inspection combined with advanced image processing technologies, such as the Faster RCNN network, for photovoltaic panel area segmentation and accurate identification of salt deposits. This method can effectively identify salt deposits on photovoltaic panels, enabling timely cleaning and maintenance. It not only helps reduce light scattering and power generation efficiency reduction caused by salt deposits, but also significantly improves the maintenance efficiency of offshore floating photovoltaic power plants, ensuring that the power plant operates in optimal condition.

[0037] This invention utilizes drones for inspections and combines them with automated image recognition technology to significantly reduce reliance on manual inspections, thereby reducing safety risks and labor costs. At the same time, this automated inspection and analysis method enables more frequent and comprehensive power plant inspections to identify and resolve problems, avoiding potential losses due to delayed maintenance.

[0038] This invention provides a wealth of accurate data support for power plant maintenance through an automated image collection and analysis process. By analyzing the data, we can better understand the patterns and characteristics of salt precipitation problems in offshore floating photovoltaic power plants, thereby optimizing maintenance processes and strategies and enabling targeted maintenance and cleaning work. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the identification method flow according to an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the salt precipitation edge marking algorithm according to an embodiment of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0043] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0044] like Figure 1-2 As shown, a method for identifying salt deposition areas of photovoltaic panels in a floating offshore photovoltaic power station includes the following steps:

[0045] S1. Drone inspection and photography of the entire photovoltaic power station, specifically:

[0046] S1-1. Based on the geographical location and layout of the offshore floating photovoltaic power station, plan the flight route of the drone to ensure that it can cover the entire area of ​​the photovoltaic power station. Considering the drone's endurance and environmental factors such as wind, reasonably arrange the flight altitude and speed to ensure safety and obtain high-quality image data.

[0047] S1-2. After the flight route planning is completed, conduct a pre-inspection of the drone, including a power check, a communication system check, and functional tests of the camera and sensors, to ensure that the drone is in good working condition.

[0048] S1-3. Following the planned route, start the drone from the designated takeoff point and fly along the predetermined route. At the same time, turn on the camera shooting function and automatically or manually control the drone's flight altitude to ensure image clarity and coverage.

[0049] S1-4. During the shooting process, adjust the flight speed and altitude of the drone according to the actual situation to adapt to the characteristics and lighting conditions of different areas, and ensure that detailed information of the photovoltaic panels and possible salt precipitation can be captured.

[0050] S1-5. After completing the entire photovoltaic power station shooting task, safely guide the drone back to the take-off point and land, and collect the image data captured by the drone for subsequent processing and analysis.

[0051] S2. Photovoltaic panel region segmentation operation to obtain an image of each photovoltaic panel, specifically including:

[0052] S2-1. The overall image of a photovoltaic power station contains numerous interfering factors, such as information beyond the photovoltaic panels, including the sea surface and floating photovoltaic structures; or the presence of multiple photovoltaic panels, affecting the detailed segmentation and identification of salt precipitation areas. Therefore, to address this issue, photovoltaic panel region segmentation is performed first. The image is then fed into a Faster R-CNN neural network trained on the photovoltaic panel images to obtain the photovoltaic panel location information. The Faster R-CNN neural network specifically includes the following components:

[0053] Data preprocessing: First, the overall photovoltaic power plant images captured by the drone are preprocessed, including adjusting the image size to meet the input requirements of the Faster R-cnn network, performing color normalization to reduce the impact of illumination changes, and applying other possible preprocessing steps (such as image enhancement) to improve the recognizability of photovoltaic panel features.

[0054] Image Input: The preprocessed image is input into the trained Faster R-CNN neural network. Faster R-CNN is a deep learning model that combines a Region Proposal Network (RPN) and Fast R-CNN, which can efficiently detect the location and category of target objects from images.

[0055] Region Proposal: In Faster RCN, a series of region proposals (i.e., image regions that may contain the target object) are first generated through the RPN. The RPN evaluates each location in the image and proposes candidate bounding boxes for regions that may contain the target (in this case, the photovoltaic panel).

[0056] Feature extraction: For each proposed region, the network extracts its features. This is done through shared convolutional layers that are capable of extracting rich feature representations from the image, which helps to identify and distinguish the photovoltaic panel from other objects (such as the sea surface, supports, etc.).

[0057] Classification and Regression: The extracted features are fed into subsequent network layers for classification and bounding box regression. The classification layer determines whether each region proposal contains a photovoltaic panel, while the regression layer fine-tunes the position of the bounding boxes to more accurately cover each photovoltaic panel.

[0058] Non-maximum suppression (NMS): Since each photovoltaic panel may be covered by multiple region proposals, the non-maximum suppression algorithm needs to be applied to eliminate proposals with excessive overlap and retain the best proposal as the final photovoltaic panel location.

[0059] Output Results: Finally, the Faster R-CNN network outputs the location information of each photovoltaic panel, including the coordinates of the bounding box. This location information can be used to crop the original image to obtain individual photovoltaic panel images for subsequent analysis and processing.

[0060] S2-2. Based on the location information of the photovoltaic panels, the original image is cropped to obtain images of all individual photovoltaic panels in the original image.

[0061] S3. The photovoltaic panel image is processed sequentially using the salt precipitation edge marking algorithm to obtain a coarse salt precipitation edge marking map, specifically including:

[0062] S3-1. To reduce noise in the photovoltaic panel image, Gaussian blurring is first applied to the image. The processing principle is as follows:

[0063]

[0064] In the formula, x is the horizontal coordinate of the image; y is the vertical coordinate of the image; P(x,y) is the pixel value at coordinate (x,y) in the processed image; I(x,y) is the pixel value at coordinate (x,y) in the original image; and σ is the preset variance value, which is mainly used to adjust the degree of Gaussian blur.

[0065] S3-2. Convert the aforementioned photovoltaic panel noise-reduced image into the HLS color space according to the following principle:

[0066]

[0067] In the formula, R, G, and B represent the pixel values ​​of the red, green, and blue channels in the RGB color space, respectively; H, L, and S represent the hue, brightness, and saturation values ​​in the HLS color space, respectively; and max and min represent the maximum and minimum values ​​of the three channels in the RGB color space, respectively.

[0068] S3-3. Next, based on the characteristics of the salt precipitation region in the HLS color space, the region is marked according to the following principle to obtain the salt precipitation region map.

[0069]

[0070] In the formula, New(x,y) is the pixel value of the processed image at (x,y); I(x,y) is the pixel value of the denoised image at (x,y); H(x,y) is the pixel value of the H channel at (x,y); L(x,y) is the pixel value of the L channel at (x,y); S(x,y) is the pixel value of the S channel at (x,y); H mean The mean value for the H channel; L mean The mean of the L channel; S mean This represents the mean of the S-channel.

[0071] S3-4. In the aforementioned salt precipitation area diagram, the salt precipitation area is filled with white. Therefore, edge recognition processing can be performed on the filled area to obtain a coarse edge map of the salt precipitation.

[0072] S4. Combine all the coarse marking maps of salt precipitation edges with the original map to identify and mark the salt precipitation areas, and obtain the salt precipitation segmentation marking map;

[0073] The standard FCN (Fully Convolutional Neural Network) takes an RGB color space image as input. However, due to the limited dataset and indistinct features in salt precipitation recognition, the standard FCN's recognition ability is poor. Therefore, the aforementioned coarse salt precipitation edge mapping is introduced, changing the original three-channel RGB color space input to a four-channel input. To preserve the original image information, the first three channels remain RGB three-channel inputs, while the fourth channel uses grayscale data from the aforementioned coarse salt precipitation edge mapping to further supplement the feature information of the salt precipitation edges, resulting in a salt precipitation region segmentation and labeling map.

[0074] Finally, a salt precipitation segmentation marking diagram was obtained for each photovoltaic panel.

[0075] S5, Image stitching;

[0076] Since the aforementioned operations are all based on individual photovoltaic panel images segmented from the original image, image stitching technology is used to stitch together all the salt precipitation area segmentation and marking images to obtain the overall photovoltaic power station's salt precipitation marking image.

[0077] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0078] This invention is intended to cover all such substitutions, modifications, and variations falling within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., 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 identifying salt deposition areas of photovoltaic panels in a floating offshore photovoltaic power station, characterized in that, Includes the following steps: S1: Use drones to take overall pictures of the floating photovoltaic power station at sea, and obtain images of the photovoltaic power station obtained by drone inspection; S2: The photovoltaic power station image is segmented by a target detection algorithm to identify and locate the position information of each photovoltaic panel. The original image is then cropped based on the position information to obtain all individual photovoltaic panel images. The target detection algorithm uses a Faster R-CNN neural network. The photovoltaic power station image is imported into the trained Faster R-CNN neural network to obtain the photovoltaic panel position regions. S3: Process the image of a single photovoltaic panel using a salt precipitation edge marking algorithm to obtain a coarse salt precipitation edge map for that panel; specifically including: S31: To perform noise reduction processing on the photovoltaic panel image; S32: Perform HLS color space conversion on the denoised image of the photovoltaic panel; S33: Based on the characteristics of the salt precipitation region in the HLS color space, perform region marking processing to obtain the salt precipitation region map; S34: In the obtained salt precipitation region map, fill the salt precipitation region with white, perform edge recognition processing on the filled region, and obtain a coarse salt precipitation edge map; The noise reduction process in S31 includes Gaussian blurring of the image, expressed as: Where x is the horizontal coordinate of the image, y is the vertical coordinate of the image, P(x,y) is the pixel value at coordinate (x,y) in the image after noise reduction, I(x,y) is the pixel value at coordinate (x,y) in the original image, and σ is the preset variance value used to adjust the degree of Gaussian blur. The HLS color space conversion in S32 is specifically represented as follows: Where R, G, and B represent the pixel values ​​of the red, green, and blue channels in the RGB color space, respectively; H, L, and S represent the hue, brightness, and saturation values ​​in the HLS color space, respectively; and max and min represent the maximum and minimum values ​​of the three channels in the RGB color space, respectively. The region marking process in S33 is as follows: Where New(x,y) is the pixel value at (x,y) in the processed image, P(x,y) is the pixel value at (x,y) in the denoised image, H(x,y) is the pixel value at (x,y) in the H channel, L(x,y) is the pixel value at (x,y) in the L channel, S(x,y) is the pixel value at (x,y) in the S channel, and H... mean The mean value for the H channel. The mean of the L channel. The mean of the S channel; S4: Combine all the coarse marking maps of salt precipitation edges with the original image to identify and mark the salt precipitation region, and obtain the salt precipitation segmentation marking map; S5: Stitch together all the salt precipitation segmentation marker images to obtain the overall salt precipitation marker image of the photovoltaic power station.

2. The method for identifying salt precipitation areas of photovoltaic panels in a floating offshore photovoltaic power station according to claim 1, characterized in that, The Faster R-CNN neural network specifically includes: Data preprocessing: The overall photovoltaic power station images captured by the drone are preprocessed, including adjusting the image size to meet the input requirements of the Faster Rcnn network and performing color normalization to reduce the impact of changes in illumination. Image input: The preprocessed image is input into the trained Faster R-CNN neural network, which combines a region proposal network and a deep learning model of R-CNN to detect the location and category of the target object from the image; Region Proposal: In Faster RCN, region proposals are generated through the RPN. The RPN evaluates each location in the image and proposes candidate bounding boxes for regions containing photovoltaic panels. Feature extraction: For each region proposal, its features are extracted through a shared convolutional layer. The convolutional layer extracts rich feature representations from the image to identify and distinguish photovoltaic panels. Classification and Regression: The extracted features are fed into subsequent network layers for classification and bounding box regression; Non-maximum suppression: The non-maximum suppression algorithm is applied to eliminate proposal boxes with excessive overlap and retain the best proposal box as the final photovoltaic panel position; Output: The Faster R-CNN neural network outputs the location information of each photovoltaic panel, including the coordinates of the bounding box. This location information is used to crop the original image to obtain individual photovoltaic panel images for subsequent analysis and processing.

3. The method for identifying salt precipitation areas of photovoltaic panels in a floating offshore photovoltaic power station according to claim 1, characterized in that, S4 includes combining the original image and all salt precipitation edge coarse marking images, and using a standard FCN fully convolutional neural network to identify and mark the salt precipitation region. The input of the FCN fully convolutional neural network is changed from the original three-channel RGB color space to a four-channel input, and the fourth channel is the grayscale data of the salt precipitation edge coarse marking image, so as to obtain the salt precipitation segmentation marking image of each photovoltaic panel.

Images