Foggy day power transformation equipment image acquisition method and system

Abstract

The invention relates to the technical field of transformers, in particular to a foggy day power transformation equipment image acquisition method and system, and the method comprises the steps: constructing a foggy power transformation equipment inspection image sample database; constructing a power transformation equipment image defogging model, wherein the power transformation equipment image defogging model comprises an image color separation module and a depth cross-scale fusion network; constructing a loss function of the power transformation equipment image defogging model, and training the power transformation equipment image defogging model by using the foggy power transformation equipment inspection image sample database; and inputting the foggy power transformation equipment image into the trained power transformation equipment image defogging model to obtain a power transformation equipment defogging image. According to the method, fog in the foggy power transformation equipment image can be efficiently removed, the image quality is improved, the quality and efficiency of power transformation equipment image acquisition in foggy weather are improved, and thus the analysis and monitoring capability of the power transformation equipment image in foggy weather is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer substations, and in particular to a method and system for collecting images of transformer substation equipment in foggy weather. Background Art

[0002] In a complex and ever-changing environment, inspection of substation equipment is crucial to ensure its normal operation. Among them, image acquisition of substation equipment in foggy scenes is particularly important.

[0003] The application of computer vision and deep learning technology in substation equipment image acquisition has significantly improved the monitoring, maintenance and management efficiency of power systems. However, although computer vision and deep learning technology have good performance in substation equipment inspection in foggy scenes, their performance in foggy environments is not satisfactory. Therefore, how to improve the acquisition effect of substation equipment images in foggy days is an urgent problem to be solved. Summary of the invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the problem in the prior art that the image acquisition effect of substation equipment is poor in foggy weather.

[0005] In order to solve the above technical problems, the present invention provides a method for collecting images of substation equipment in foggy weather, comprising:

[0006] Collect inspection images of foggy substation equipment and build an original foggy image database; preprocess the original foggy image database and expand the samples to obtain a foggy substation equipment inspection image sample database;

[0007] Construct a substation image defogging model; the substation image defogging model includes an image color separation module and a deep cross-scale fusion network; the image color separation module separates the input foggy substation image to obtain a red channel image, a green channel image and a blue channel image, each channel image is sequentially passed through a convolution layer and an activation function and then spliced, and the spliced ​​image is sequentially passed through a convolution layer and an activation function to obtain the output image of the image color separation module; the output image of the image color separation module is input into the deep cross-scale fusion network to output a predicted fog image; the input foggy substation image is subtracted from the predicted fog image to obtain a defogging image of the substation;

[0008] The loss function of the substation image defogging model is constructed and the network parameters are initialized. The substation image defogging model is trained using the foggy substation inspection image sample database to obtain a trained substation image defogging model.

[0009] The foggy substation equipment image is input into the trained substation equipment image defogging model to obtain the defogging image of the substation equipment.

[0010] Preferably, the preprocessing of the original foggy image database and sample expansion to obtain a foggy substation equipment inspection image sample database includes:

[0011] The foggy substation inspection images in the original foggy image database are standardized and adjusted to a uniform size.

[0012] Gaussian denoising is performed on the standardized inspection images of foggy substation equipment;

[0013] Image enhancement is performed on the denoised inspection image of foggy substation equipment.

[0014] Preferably, the image enhancement operation includes horizontal flipping, vertical flipping, brightness adjustment, blurring and random grayscale transformation.

[0015] Preferably, each channel image is spliced ​​after passing through a convolution layer and an activation function in sequence, and the spliced ​​image is spliced ​​after passing through a convolution layer and an activation function in sequence to obtain an output image of the image color separation module, including:

[0016] The red channel image passes through the first convolution layer and the Relu activation function to obtain the red channel feature map; the green channel image passes through the second convolution layer and the Relu activation function to obtain the green channel feature map; the blue channel image passes through the third convolution layer and the Relu activation function to obtain the blue channel feature map;

[0017] The red channel feature map, the green channel feature map and the blue channel feature map are spliced ​​to obtain a spliced ​​image; the spliced ​​image is sequentially passed through the fourth convolutional layer and the activation function to obtain the output image of the image color separation module.

[0018] Preferably, the formulas of the first convolutional layer, the second convolutional layer, the third convolutional layer and the fourth convolutional layer are expressed as:

[0019] F c =W*I c +A,c∈[R,G,B]

[0020] Among them, F c is the output of the convolution layer, W and A represent the weight coefficient matrix and bias matrix of the convolution respectively, * is the convolution operation, I c Represents the convolution layer input, R represents the red channel, G represents the green channel, and B represents the blue channel; the stride of all convolution operations is set to 1.

[0021] Preferably, the number of convolution kernels of the first convolution layer, the second convolution layer and the third convolution layer are all 3, and the size of the convolution kernels is 3×3×1; the number of convolution kernels of the fourth convolution layer is 9, and the size of the convolution kernel is 3×3×9.

[0022] Preferably, the maximum pooling operation in the deep cross-scale fusion network is replaced by an average pooling operation.

[0023] Preferably, the standard convolution in the deep cross-scale fusion network is replaced with depthwise separable convolution.

[0024] Preferably, the SSIM loss function is used as the loss function of the substation image defogging model.

[0025] The present invention also provides an image acquisition system for substation equipment in foggy weather, comprising:

[0026] The database construction module is used to collect inspection images of foggy substation equipment and build an original foggy image database; preprocess the original foggy image database and perform sample expansion to obtain a foggy substation equipment inspection image sample database;

[0027] A model building module is used to build a substation image defogging model; the substation image defogging model includes an image color separation module and a deep cross-scale fusion network; the image color separation module separates the input foggy substation image to obtain a red channel image, a green channel image and a blue channel image, each channel image is respectively passed through a convolution layer and an activation function in turn and then spliced, and the spliced ​​images are passed through a convolution layer and an activation function in turn to obtain an output image of the image color separation module; the output image of the image color separation module is input into the deep cross-scale fusion network to output a predicted fog image; the input foggy substation image is subtracted from the predicted fog image to obtain a defogging image of the substation;

[0028] The training module is used to construct the loss function of the substation image defogging model and initialize the network parameters, and train the substation image defogging model using the foggy substation inspection image sample database to obtain a trained substation image defogging model;

[0029] The defogging module is used to input the foggy substation equipment image into the trained substation equipment image defogging model to obtain the defogging image of the substation equipment.

[0030] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0031] The method for collecting images of substation equipment in foggy weather described in the present invention constructs a substation equipment image defogging model, and takes a deep cross-scale fusion network as a basis to effectively extract and fuse the features of the foggy substation equipment image. In addition, an image color separation module is introduced before the deep cross-scale fusion network to independently extract the RGB three-channel color features of the foggy substation equipment image, thereby improving the performance of the substation equipment image defogging model, being able to efficiently remove fog from the foggy substation equipment image and improve the image quality, thereby improving the quality and efficiency of substation equipment image collection in foggy weather, and thus improving the analysis and monitoring capabilities of the foggy substation equipment image.

[0032] The method for acquiring images of substation equipment in foggy weather described in the present invention uses an average pooling operation to replace the maximum pooling operation in a deep cross-scale fusion network, aiming to balance the globality and locality of image features, so that the substation equipment image defogging model can more effectively handle image noise and detail loss when removing fog, thereby improving the defogging effect.

[0033] In addition, in order to reduce the computational complexity of the substation image defogging model and maintain the accuracy and efficiency of the defogging effect, the present invention also introduces a deep separable convolution to replace the standard convolution layer in the deep cross-scale fusion network. The deep separable convolution reduces the computational load of the model and maintains the accuracy and efficiency of the defogging effect by separating the convolution operations between channels and in space. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0035] Figure 1 It is a flow chart of a method for collecting images of power substation equipment in foggy weather according to the present invention;

[0036] Figure 2 It is a structural diagram of the image color separation module;

[0037] Figure 3 It is a schematic diagram of the encoder-decoder structure of the deep cross-scale fusion network;

[0038] Figure 4 It is a schematic diagram of the internal connection module of the deep cross-scale fusion network;

[0039] Figure 5 This is a structural comparison diagram of depth-wise separable convolution and standard convolution. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0041] Reference Figure 1 As shown, the present invention provides a method for collecting images of substation equipment in foggy weather, which specifically includes:

[0042] S1. Collect inspection images of foggy substation equipment and build an original foggy image database; preprocess the original foggy image database and expand the samples to obtain a foggy substation equipment inspection image sample database;

[0043] S2. Construct a substation image defogging model; the substation image defogging model includes an image color separation module and a deep cross-scale fusion network; the image color separation module separates the input foggy substation image into a red channel image, a green channel image and a blue channel image, each channel image is sequentially passed through a convolution layer and an activation function and then spliced, and the spliced ​​image is sequentially passed through a convolution layer and an activation function to obtain the output image of the image color separation module; the output image of the image color separation module is input into the deep cross-scale fusion network to output a predicted fog image; the input foggy substation image is subtracted from the predicted fog image to obtain a defogging image of the substation;

[0044] S3, construct the loss function of the substation image defogging model and initialize the network parameters, use the foggy substation inspection image sample database to train the substation image defogging model, and obtain the trained substation image defogging model;

[0045] S4. Input the foggy substation equipment image into the trained substation equipment image defogging model to obtain the defogging image of the substation equipment.

[0046] Specifically, S1 includes the following steps:

[0047] S101. Collection and acquisition of inspection images of foggy substation equipment.

[0048] This embodiment continuously collects inspection images of foggy substation equipment through the intelligent inspection device deployed on site. The collected inspection images of foggy substation equipment will be remotely transmitted to the data center through the network transmission protocol and stored in the remote server to build the original foggy image database. In this process, a large number of foggy substation equipment inspection image samples accumulated continuously will be used for the training of subsequent image processing algorithms to improve the image processing accuracy of the algorithm in foggy environments.

[0049] S102, preprocessing the original foggy image database and performing sample expansion to obtain a foggy substation equipment inspection image sample database.

[0050] Firstly, the foggy substation inspection images in the original foggy image database are standardized and the image size is adjusted to a uniform 640x640 pixels. Then, the Gaussian denoising algorithm is implemented on the foggy substation inspection images to reduce the impact of environmental noise on image quality.

[0051] In order to further enhance the diversity and robustness of the image sample database, a series of image enhancement operations are performed on the inspection images of foggy substation equipment, including but not limited to horizontal flipping, vertical flipping, brightness adjustment (including simulation of low-brightness and high-brightness scenes), blur processing, and random grayscale transformation. These enhancement operations help improve the recognition ability of subsequent image processing algorithms for image features under various environmental conditions.

[0052] Through the above steps, the present invention can effectively process and analyze substation equipment inspection images collected under complex environmental conditions, especially under foggy meteorological conditions, thereby improving the accuracy and reliability of the inspection.

[0053] Specifically, the substation image defogging model is constructed in S2, including the following steps:

[0054] S201, introducing an image color separation module.

[0055] Reference Figure 2 As shown, the image color separation module separates the input image into a red channel image, a green channel image and a blue channel image. Each channel image is a single-channel grayscale image. For clear display, this embodiment restores each channel image to a three-channel image.

[0056] Each channel image passes through a convolution layer and an activation function in turn to obtain the feature map of each channel. Specifically, the red channel image passes through the first convolution layer and the Relu activation function to obtain the red channel feature map; the green channel image passes through the second convolution layer and the Relu activation function to obtain the green channel feature map; the blue channel image passes through the third convolution layer and the Relu activation function to obtain the blue channel feature map.

[0057] The feature maps of each channel are spliced ​​to obtain a spliced ​​image. The spliced ​​image passes through the fourth convolution layer and activation function in sequence to obtain the output image of the image color separation module.

[0058] The first convolution layer, the second convolution layer, the third convolution layer and the fourth convolution layer are all standard convolutions, and the formula is expressed as:

[0059] F c =W*I c +A,c∈[R,G,B]

[0060] Among them, F cis the output of the convolution layer, W and A represent the weight coefficient matrix and bias matrix of the convolution layer respectively, * is the convolution operation, I c represents the convolution layer input, R represents the red channel, G represents the green channel, and B represents the blue channel. The step size of all convolution operations is set to 1. The specific convolution layer information is shown in Table 1.

[0061] Table 1. Information of each convolutional layer of the image color separation module

[0062] type Input size Convolution kernel size Number of convolution kernels The first convolutional layer 1024×1024×1 3×3×1 3 The second convolutional layer 1024×1024×1 3×3×1 3 The third convolutional layer 1024×1024×1 3×3×1 3 The fourth convolutional layer 1024×1024×9 3×3×9 9

[0063] The image color separation module uses a series of convolutional layers and activation function processing steps to help extract key information and features from the image. By splicing multi-channel feature maps and further convolution operations, the image acquisition system can enhance its ability to process and analyze images of foggy substation equipment, which helps improve the performance and accuracy of the image acquisition system.

[0064] After the foggy image is divided into RGB three-channel images, each channel image contains fog of corresponding density. In the original image, the RGB channel image corresponding to the area with higher fog density also has higher fog density. In order to obtain a more accurate haze concentration map, this embodiment introduces an image color separation module into the substation image defogging model to extract the shallow haze concentration features of different color channel images.

[0065] S202. Build a deep cross-scale fusion network.

[0066] The Difficulties in Capturing Sharp Foggy Network (DCSFN) structure contains multiple sub-network structures, and each sub-network extracts image information by learning features of different scales. In order to fuse feature information of different scales, this embodiment adopts the method of gated recurrent unit (GRU). Encoder modules and decoder modules are designed in each sub-network of the system. Dense connections are achieved within the codec module and between codecs through jump connections to facilitate feature propagation and reuse. After the input image passes through the codec, a feature fusion operation is performed to fuse the features learned by different sub-networks to obtain a predicted fog image. The predicted fog image is subtracted from the input foggy substation image to obtain a defogged image of the substation.

[0067] First, the input image is preprocessed by a 3×3 convolution operation, and then the image is input into sub-networks of different scales through a maximum pooling operation (downsampling). Each sub-network consists of an encoder module and a decoder module, and this structure is the same within each sub-network.

[0068] Reference Figure 3 As shown in the figure, inside the encoder module, different encoders are densely connected through jump connections, and the decoder module also uses this connection method. In addition, the corresponding encoders and decoders are also densely connected through jump connections. The introduction of dense connection structure has multiple benefits. It can alleviate the gradient vanishing problem, enhance the feature propagation capability, and realize feature reuse, so as to better predict the output of the network. At the same time, dense connection can also speed up the training of the network. The purpose of adding GRU between the decoder and the encoder is to enable GRU to process the feature map generated by the encoder, and then pass the processed features to the decoder to help the decoder better reconstruct the image.

[0069] Inside each codec, there are also densely connected branches of different scales, as shown in the following formula:

[0070]

[0071] in, and are the input features of the first, second, third and Lth encoders of the sub-network respectively, is the output of the l-th encoder, L is the total number of encoders, C[·] is the unified operation of different features on the number of channels, ξ(·) is the feature fusion operation, and the convolution with a convolution kernel size of 1×1 is used; τ(·) is the internal connection module.

[0072] Refer to the structural diagram of the internal connection module Figure 4 As shown in the figure. The encoder inside the system is divided into branches with different scales. The input features are first subjected to the maximum pooling operation to obtain multi-scale features and form multiple branches. Next, each branch undergoes multiple convolution operations and then uses a jump connection to perform feature fusion operations. This improved method uses a combination of multi-scale features and jump connections to enhance the effect of collecting substation images in foggy environments. Through the maximum pooling operation, the system is able to obtain feature information of different scales, thereby fully capturing the details of the substation. In each branch, multiple convolution operations further process the features to extract more useful information. Finally, the jump connection fuses the features of each branch to obtain a more global and comprehensive representation.

[0073] The deep cross-scale fusion network achieves dense connection by applying jump connections multiple times, and its main purpose is to predict fog images more accurately. The deep cross-scale fusion network aims to better predict the fog concentration map of foggy input images. Applying the deep cross-scale fusion network to the defogging task of substation images can effectively collect images of substations in foggy weather and provide clearer and more accurate results. The deep cross-scale fusion network uses complex image processing technology and architecture design, combined with professional field knowledge, which helps to improve the acquisition quality and visibility of substation images in foggy weather, realizes accurate prediction of fog concentration maps, and provides an efficient and reliable solution for the acquisition and analysis of substation images in foggy weather. By optimizing the network model and algorithm, the present invention brings new possibilities for the development and application of substation image acquisition systems in foggy weather, and has broad application prospects in the field of image defogging.

[0074] S203, constructing a substation image defogging model, wherein the substation image defogging model includes an image color separation module and a deep cross-scale fusion network connected in sequence.

[0075] Preferably, the present invention optimizes the deep cross-scale fusion network of the substation image defogging model, including:

[0076] S204, replacing the maximum pooling operation in the deep cross-scale fusion network with an average pooling operation.

[0077] When it comes to substation image acquisition in a foggy environment, the presence of haze and atmospheric particulate matter greatly affects the clarity and visibility of the image. This embodiment further improves the system performance and image quality through pooling operations.

[0078] The core goal of the pooling operation is to reduce redundant information, reduce the amount of calculation, and prevent overfitting while retaining key information. In order to achieve this goal, average pooling and maximum pooling have become common technical means.

[0079] Average pooling calculates the average of the pixel values ​​in the image area and uses it as the value of the next layer of elements. This process can effectively reduce the redundant information in the image, allowing the system to focus more on the main features and content of the image. On the other hand, maximum pooling achieves abstraction and retention of key features of the image by selecting the maximum pixel value in the image area. This operation can highlight important visual information in foggy environments and improve the recognition and comprehensibility of images.

[0080] The average pooling and maximum pooling operations are shown below:

[0081]

[0082] According to the above formula, average pooling calculates the average pixel value of each area in the feature map to reduce redundant information and retain key information. This operation allows the system to focus more on the main features and content of the image, thereby improving the accuracy of image acquisition. On the other hand, maximum pooling highlights the key visual information in foggy environments by selecting the maximum pixel value of each area in the feature map.

[0083] Depending on the different situations of the feature map, this embodiment selects the pooling operation according to the following principles: when most of the information in the feature map is important, average pooling is selected to retain the background information; when only part of the information in the feature map is obviously important, maximum pooling is selected to retain the main features.

[0084] Considering that haze in foggy images usually forms regional rather than point distribution, in this embodiment, the maximum pooling operation in the deep cross-scale fusion network is replaced by an average pooling operation, which can better predict the haze concentration distribution in the input foggy image, thereby improving the analysis and monitoring capabilities of foggy substation equipment images.

[0085] The optimized substation equipment image dehazing model can effectively process substation equipment images in foggy environments, accurately capture important information, and reduce the impact of factors such as haze and particulate matter. It not only improves the accuracy and efficiency of image acquisition, but also provides more reliable data support for the monitoring and maintenance of substation equipment.

[0086] S205. Replace the standard convolution in the deep cross-scale fusion network with a depth-wise separable convolution.

[0087] This embodiment introduces the depth-separable convolution technology, which includes channel-by-channel convolution and point-by-point convolution. Compared with the traditional standard convolution, the depth-separable convolution has only made minor modifications, but it significantly reduces the number of network parameters, thereby reducing the computational complexity and improving the efficiency of the network.

[0088] To better explain the advantages of depthwise separable convolution, we first introduce the traditional standard convolution. Assume that the size of the input image is 5×5×3, the convolution kernel size is set to 3, and the number of channels of the output feature map is 4. According to the calculation method of standard convolution, the number of convolution kernels required is 4, and the number of channels of each convolution kernel is 3. Therefore, the total number of parameters is 3×3×3×4=108. When the stride is 1, the size of the output image after the convolution operation is 5×5. If no padding is performed, the output feature map size is 3×3 (same as padding). In summary, standard convolution resizes the feature map of the previous layer by adjusting the convolution kernel and stride size.

[0089] However, depthwise separable convolution includes channel-by-channel convolution and point-by-point convolution. The channel-by-channel convolution first performs a separate convolution calculation on each channel of the input to generate an intermediate feature map with the same number of channels. Then, the point-by-point convolution fuses the intermediate feature maps obtained by the channel-by-channel convolution pixel by pixel to finally obtain the output feature map. Through such operations, the depthwise separable convolution reduces the number of parameters from 108 to 3×3×1×4+1×1×4=40, greatly reducing the number of parameters of the network. For a structural comparison of the depthwise separable convolution and the standard convolution, refer to Figure 5 shown.

[0090] The advantage of the depthwise separable convolution is its efficient parameter utilization and computational performance. By reducing the number of parameters, the amount of computation of the network is significantly reduced, thereby improving the computational efficiency of the system. At the same time, the depthwise separable convolution can effectively extract useful feature information while maintaining fewer parameters, which is of great significance for the acquisition and processing of images of substation equipment in foggy weather.

[0091] Based on the above description, the optimized substation equipment image dehazing model takes deep separable convolution as the core technology. Through reasonable parameter design and calculation strategy, it effectively improves the performance and efficiency of the model, and has significant advantages in the collection and analysis of substation equipment images in foggy environments.

[0092] The improvements of S204 and S205 can be applied separately to deep cross-scale fusion networks.

[0093] Specifically, S3 includes the following steps:

[0094] S301, constructing the loss function of the substation image defogging model and initializing the network parameters.

[0095] This embodiment uses the structural similarity index (SSIM) as the loss function for training the substation image defogging model. SSIM is an indicator that measures the structural similarity between two images. It combines the differences in brightness, contrast, and structure, and can provide better results in image quality assessment. SSIM is selected as the loss function because it can effectively retain the structural information of the image in the process of improving the defogging algorithm and reduce the distortion and noise in the substation image in foggy weather.

[0096] During the training process of the system, this embodiment uses the SSIM loss function as the objective function and optimizes the network parameters through the back propagation algorithm. Specifically, the SSIM between the real foggy substation image and the defogging blurred image is used as the loss value of the training sample, and the performance of the network is improved by minimizing this loss function. The advantage of using the SSIM loss function is that the weights can be adaptively adjusted according to the structural similarity of the image, thereby better meeting the requirements of the defogging algorithm.

[0097] S302: Introduce batch normalization technology.

[0098] In addition, in order to improve the stability and robustness of the system, we also introduced batch normalization technology to normalize the input data distribution between network layers to reduce the occurrence of gradient vanishing and gradient exploding problems. Batch normalization technology can effectively accelerate the convergence process of the network and improve the generalization ability of the network during training.

[0099] In summary, this embodiment uses the SSIM loss function in the training process and introduces batch normalization technology to optimize network parameters, improve the performance of the defogging algorithm, and maintain the structural information of the image. This innovative method is of great significance for improving the image quality of substation equipment in foggy environments.

[0100] S303: train a substation image defogging model using a foggy substation inspection image sample database to obtain a trained substation image defogging model.

[0101] In the training phase, this embodiment uses a large number of real-life foggy substation equipment images and their corresponding clear image label data for network training. By using the labeled clear images as references and combining the structural similarity index mentioned in S301 as the loss function, the model parameters are optimized so that the substation equipment image defogging model can accurately restore the details of the clear image and effectively remove the fog component.

[0102] Specifically, during the training process, the original foggy substation images are first preprocessed and data enhanced to increase the diversity and quantity of training data. Then, by extracting image features and fusing information at different levels, the substation image defogging model can learn the fog distribution and image details in the substation images.

[0103] Based on the structural similarity index, this embodiment also compares the clear image and the defogging image generated by the substation image defogging model, and calculates the loss based on the difference between them. Specifically, the structural similarity index is based on the contrast measurement of brightness, contrast and structure, and can quantify the similarity between images. By minimizing the loss of the structural similarity index, the network model can better fit the clear image of real visual perception, thereby achieving the accuracy and reliability of image defogging.

[0104] Therefore, in the training phase of the present invention, by using a large amount of data from real scenes and the structural similarity index as the loss function, the substation image defogging model has the ability to accurately restore clear image details and effectively remove fog. Through this optimization process, the image acquisition method of substation equipment in foggy weather provided by the present invention can provide higher quality image information support for applications such as substation equipment inspection.

[0105] In summary, the present invention uses the substation image defogging model to achieve defogging of substation images in foggy weather, producing higher-definition image results, which has important application value in image acquisition and processing in foggy environments.

[0106] This embodiment further uses Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) as indicators to compare the method of the present invention with the prior art to illustrate the beneficial effects of the method of the present invention.

[0107] Peak signal-to-noise ratio is often widely used in image quality evaluation as the main indicator. The calculation of PSNR is based on mean square error (MSE). Generally speaking, the larger the PSNR value, the better the image quality.

[0108] Structural similarity is a metric used to measure the similarity between two images. When evaluating the dehazing effect, an image processed by a dehazing algorithm is usually compared with the corresponding original haze-free image. Generally speaking, a higher SSIM value means that the restored haze-free image successfully retains more structural information, indicating that its quality is better. The similarity of images is mainly evaluated from three aspects: structure, brightness, and contrast.

[0109] The larger the values ​​of the objective evaluation indicators Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM), the better the quality of the restored image. This means that the dehazing algorithm effectively improves the clarity and detail restoration capabilities of the image, resulting in a significant improvement in the quality of the dehazed image.

[0111] Table 2. Comparison of the results of the substation image defogging model of the present invention and the existing defogging algorithm

[0112]

[0113] Analyzing the data in Table 2, the PSNR (27.81) of the present invention is significantly better than other comparison algorithms, especially compared with the DCP algorithm (12.94) and the AOD-Net algorithm (17.90). This shows that the present invention can better restore image details and maintain image clarity during the defogging process, thereby significantly improving image quality.

[0114] In addition, the SSIM (0.9391) of the present invention is close to that of the MSBDN algorithm (0.9023) and is superior to other algorithms, indicating that the present invention can better preserve the structural information, brightness and contrast of the original image during the defogging process, so that the restored haze-free image has higher quality.

[0115] In summary, through experimental data comparison, the present invention shows good performance under the two objective evaluation indicators of peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), and has achieved significant improvement in image quality compared with other comparison algorithms.

[0116] Based on the above-mentioned image acquisition method for a power substation in foggy weather, this embodiment further provides an image acquisition system for a power substation in foggy weather, including:

[0117] The database construction module is used to collect inspection images of foggy substation equipment and build an original foggy image database; preprocess the original foggy image database and perform sample expansion to obtain a foggy substation equipment inspection image sample database;

[0118] A model building module is used to build a substation image defogging model; the substation image defogging model includes an image color separation module and a deep cross-scale fusion network; the image color separation module separates the input foggy substation image to obtain a red channel image, a green channel image and a blue channel image, each channel image is respectively passed through a convolution layer and an activation function in turn and then spliced, and the spliced ​​images are passed through a convolution layer and an activation function in turn to obtain an output image of the image color separation module; the output image of the image color separation module is input into the deep cross-scale fusion network to output a predicted fog image; the input foggy substation image is subtracted from the predicted fog image to obtain a defogging image of the substation;

[0119] The training module is used to construct the loss function of the substation image defogging model and initialize the network parameters, and train the substation image defogging model using the foggy substation inspection image sample database to obtain a trained substation image defogging model;

[0120] The defogging module is used to input the foggy substation equipment image into the trained substation equipment image defogging model to obtain the defogging image of the substation equipment.

[0121] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0122] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0123] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0124] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0125] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A method for collecting images of power substation equipment in foggy weather, characterized in that: include: Collect inspection images of foggy substation equipment and build an original foggy image database; The original foggy image database is preprocessed and sample expanded to obtain a foggy substation equipment inspection image sample database; Construct a substation image defogging model; the substation image defogging model includes an image color separation module and a deep cross-scale fusion network; the image color separation module separates the input foggy substation image to obtain a red channel image, a green channel image and a blue channel image, each channel image is sequentially passed through a convolution layer and an activation function and then spliced, and the spliced ​​image is sequentially passed through a convolution layer and an activation function to obtain the output image of the image color separation module; the output image of the image color separation module is input into the deep cross-scale fusion network to output a predicted fog image; the input foggy substation image is subtracted from the predicted fog image to obtain a defogging image of the substation; The loss function of the substation image defogging model is constructed and the network parameters are initialized. The substation image defogging model is trained using the foggy substation inspection image sample database to obtain a trained substation image defogging model. The foggy substation equipment image is input into the trained substation equipment image defogging model to obtain the defogging image of the substation equipment.

2. The method for collecting images of power substation equipment in foggy weather according to claim 1, characterized in that: The original foggy image database is preprocessed and sample expanded to obtain a foggy substation equipment inspection image sample database, including: The foggy substation inspection images in the original foggy image database are standardized and adjusted to a uniform size. Gaussian denoising is performed on the standardized inspection images of foggy substation equipment; Image enhancement is performed on the denoised inspection image of foggy substation equipment.

3. The method for collecting images of power substation equipment in foggy weather according to claim 2, characterized in that: The image enhancement operations include horizontal flipping, vertical flipping, brightness adjustment, blurring, and random grayscale transformation.

4. The method for collecting images of power substation equipment in foggy weather according to claim 1, characterized in that: Each channel image is sequentially passed through a convolution layer and an activation function and then spliced, and the spliced ​​image is sequentially passed through a convolution layer and an activation function to obtain an output image of the image color separation module, including: The red channel image passes through the first convolutional layer and the Relu activation function to obtain the red channel feature map; The green channel image passes through the second convolutional layer and the Relu activation function to obtain the green channel feature map; The blue channel image passes through the third convolutional layer and the Relu activation function to obtain the blue channel feature map; The red channel feature map, the green channel feature map and the blue channel feature map are spliced ​​to obtain a spliced ​​image; the spliced ​​image is sequentially passed through the fourth convolutional layer and the activation function to obtain the output image of the image color separation module.

5. The method for collecting images of power substation equipment in foggy weather according to claim 4, characterized in that: The formulas of the first convolutional layer, the second convolutional layer, the third convolutional layer and the fourth convolutional layer are expressed as: F c =W*I c +A,c∈[R,G,B] Among them, F c is the output of the convolution layer, W and A represent the weight coefficient matrix and bias matrix of the convolution respectively, * is the convolution operation, I c Represents the convolution layer input, R represents the red channel, G represents the green channel, and B represents the blue channel; the stride of all convolution operations is set to 1.

6. The method for collecting images of power substation equipment in foggy weather according to claim 4, characterized in that: The number of convolution kernels of the first convolution layer, the second convolution layer and the third convolution layer are all 3, and the size of the convolution kernel is 3×3×1; the number of convolution kernels of the fourth convolution layer is 9, and the size of the convolution kernel is 3×3×9.

7. The method for collecting images of power substation equipment in foggy weather according to claim 1, characterized in that: The maximum pooling operation in the deep cross-scale fusion network is replaced by the average pooling operation.

8. The method for collecting images of power substation equipment in foggy weather according to claim 1, characterized in that: Replace the standard convolution in the deep cross-scale fusion network with depthwise separable convolution.

9. The method for collecting images of power substation equipment in foggy weather according to claim 1, characterized in that: The SSIM loss function is used as the loss function of the substation image dehazing model.

10. A foggy substation equipment image acquisition system, characterized in that: include: Database construction module, used to collect inspection images of foggy substation equipment and build an original foggy image database; The original foggy image database is preprocessed and sample expanded to obtain a foggy substation equipment inspection image sample database; A model building module is used to build a substation image defogging model; the substation image defogging model includes an image color separation module and a deep cross-scale fusion network; the image color separation module separates the input foggy substation image to obtain a red channel image, a green channel image and a blue channel image, each channel image is sequentially passed through a convolution layer and an activation function and then spliced, and the spliced ​​image is sequentially passed through a convolution layer and an activation function to obtain the output image of the image color separation module; the output image of the image color separation module is input into the deep cross-scale fusion network, Output the predicted fog image; subtract the predicted fog image from the input foggy substation equipment image to obtain the defogging image of the substation equipment; The training module is used to construct the loss function of the substation image defogging model and initialize the network parameters, and train the substation image defogging model using the foggy substation inspection image sample database to obtain a trained substation image defogging model; The defogging module is used to input the foggy substation equipment image into the trained substation equipment image defogging model to obtain the defogging image of the substation equipment.

Images