Thermal power station remote sensing detection method and system based on adaptive Fourier convolution

Abstract

The invention relates to a thermal power station remote sensing detection method and system based on adaptive Fourier convolution. The method comprises the steps of collecting a remote sensing image of a thermal power station area through a remote sensing technology, performing data preprocessing to obtain a processed remote sensing image, inputting the processed remote sensing image into an encoder structure, and processing the processed remote sensing image through a self-adaptive Fourier convolution layer in the encoder structure, the method comprises the steps of obtaining a feature map, inputting the feature map into a decoder structure, reconstructing and recognizing exhaust features of a cooling tower in the feature map through the decoder structure, outputting a classification result, and determining the working state of the thermal power station. On the basis of mutual cooperation of a Fourier domain pooling layer, an asymmetric convolutional layer, a feature fusion part, a self-adaptive multi-scale fusion layer and a patch expansion part, high-precision detection can be kept in a complex environment, the recognition capability of the exhaust condition of a cooling tower is improved, and therefore the working state of a thermal power station is accurately and efficiently judged.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a remote sensing detection method and system for a thermal power station based on adaptive Fourier convolution. Background Art

[0002] With the development of society, smart grid, as the core pillar in the grand blueprint of global energy Internet, has a solid foundation in the accurate grasp of the distribution of thermal power plants around the world and their detailed spatial information. As the mainstay of the energy supply system, the real-time monitoring of the operating status of thermal power plants is not only related to the refinement of energy management, but also an important consideration for environmental protection and sustainable development strategies. Thermal power generation occupies a dominant position in the power production of many countries. Thermal power plants go through a complex energy conversion process, from the combustion of fossil fuels to the propulsion of steam, and then to the generation of electricity. Each step is accompanied by potential impacts on the environment and energy efficiency considerations. However, in the face of increasingly stringent environmental protection standards and energy efficiency requirements, efficient and accurate monitoring and management of the operating status of thermal power plants has become a global issue that needs to be solved urgently. The recycling mechanism of cooling towers, although it embodies the wisdom of resource reuse, also highlights the complexity of system monitoring and management. In order to strengthen the monitoring of thermal power plants, remote sensing images of thermal power plants are often collected, and then monitoring is achieved through feature extraction; or manual inspections and fixed sensors are relied on to collect the operating data of thermal power plants.

[0003] However, in remote sensing image processing with complex backgrounds and multi-scale changes, traditional methods are limited by noise and interference, making it difficult to accurately extract key features of thermal power plants; existing thermal power plant monitoring mainly relies on manual inspections and fixed sensors, which have problems such as low efficiency, limited coverage, and poor real-time performance. Therefore, traditional detection methods are difficult to achieve automated and accurate monitoring over a large area, and have problems with low detection accuracy and efficiency. Summary of the invention

[0004] Based on this, in order to solve the above technical problems, a remote sensing detection method and system for a thermal power station based on adaptive Fourier convolution is provided, which can improve the detection accuracy and efficiency.

[0005] A remote sensing detection method for a thermal power station based on adaptive Fourier convolution, the method comprising:

[0006] Collecting remote sensing images of the thermal power station area through remote sensing technology, and performing data preprocessing operations on the remote sensing images to obtain processed remote sensing images;

[0007] Inputting the processed remote sensing image into an encoder structure, and processing the processed remote sensing image through an adaptive Fourier convolution layer in the encoder structure to obtain a feature map;

[0008] Inputting the feature map into a decoder structure, reconstructing and identifying the exhaust characteristics of the cooling tower in the feature map through the decoder mechanism, and outputting a classification result;

[0009] determining the working state of the thermal power station according to the classification result;

[0010] Among them, the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part; the decoder mechanism is provided with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part.

[0011] In one embodiment, performing data preprocessing on the remote sensing image to obtain a processed remote sensing image includes:

[0012] Using a filter to remove noise from the remote sensing image to obtain a remote sensing image after noise removal;

[0013] The remote sensing image after noise removal is corrected by radiation correction to obtain a corrected remote sensing image;

[0014] The corrected remote sensing image is geometrically corrected to obtain a processed remote sensing image.

[0015] In one embodiment, removing noise from the remote sensing image using a filter comprises:

[0016] The filter is used to remove the noise in the remote sensing image by means of median filtering, mean filtering or convolution filtering.

[0017] In one embodiment, the step of correcting the remote sensing image after noise removal by radiation correction to obtain the corrected remote sensing image includes:

[0018] Obtain the received radiation value through the sensor and determine the actual radiation value on the ground;

[0019] The received radiation value is compared with the actual ground radiation value, and based on the comparison result, the remote sensing image after noise removal is corrected to obtain a corrected remote sensing image.

[0020] In one embodiment, the processing of the processed remote sensing image by the adaptive Fourier convolution layer in the encoder structure to obtain a feature map includes:

[0021] Performing Fourier transform on the processed remote sensing image through the Fourier domain pooling layer to extract spectral features in the processed remote sensing image;

[0022] Performing an asymmetric convolution operation on the processed remote sensing image through the first asymmetric convolution layer to obtain directional features in the processed remote sensing image;

[0023] The spectral features and directional features are fused through the feature fusion part, and processed through batch normalization and activation function to obtain a feature map.

[0024] In one of the embodiments, a plurality of the adaptive Fourier convolution layers are provided in the encoder structure, and the number of channels in each of the adaptive Fourier convolution layers increases sequentially.

[0025] In one embodiment, the reconstructing and identifying the exhaust characteristics of the cooling tower in the characteristic map by the decoder mechanism and outputting the classification result includes:

[0026] Applying an adaptive multi-scale fusion technology to the feature map through the adaptive multi-scale fusion layer to capture multi-scale features, and fusing the multi-scale features;

[0027] The fused multi-scale features are passed through the second asymmetric convolutional layer and the patch expansion part to perform feature reconstruction and recognition to obtain a classification result.

[0028] In one of the embodiments, the adaptive multi-scale fusion layer and the second asymmetric convolution layer perform feature processing on the feature map alternately;

[0029] The patch expansion part is arranged after the second asymmetric convolution layer, and gradually restores the spatial resolution of the image through upsampling operation to complete the feature image reconstruction.

[0030] In one of the embodiments, a meta-attention mechanism module is provided after the adaptive Fourier convolution layer and in the decoder mechanism for performing feature weighting.

[0031] A remote sensing detection system for a thermal power station based on adaptive Fourier convolution, the system comprising:

[0032] An image acquisition and processing module is used to acquire remote sensing images of the thermal power station area through remote sensing technology, and perform data preprocessing operations on the remote sensing images to obtain processed remote sensing images;

[0033] An encoder processing module, used for inputting the processed remote sensing image into an encoder structure, processing the processed remote sensing image through an adaptive Fourier convolution layer in the encoder structure to obtain a feature map;

[0034] A decoder processing module, used for inputting the feature map into a decoder structure, reconstructing and identifying the exhaust characteristics of the cooling tower in the feature map through the decoder mechanism, and outputting a classification result;

[0035] A state determination module, used to determine the working state of the thermal power station according to the classification result;

[0036] Among them, the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part; the decoder mechanism is provided with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part.

[0037] The above-mentioned thermal power station remote sensing detection method and system based on adaptive Fourier convolution is based on the cooperation of Fourier domain pooling layer, asymmetric convolution layer, feature fusion part, adaptive multi-scale fusion layer, and patch expansion part. When extracting image features, it can efficiently learn and extract fine-grained spectral features in the image, effectively remove noise and interference; flexibly capture directional features and multi-scale information in the image, and improve the accuracy and efficiency of feature extraction; among them, multi-scale feature extraction helps to maintain high-precision detection in complex environments, improve the ability to recognize the exhaust conditions of cooling towers, and can accurately identify the cooling towers and their exhaust characteristics in remote sensing images, so as to accurately and efficiently judge the working status of thermal power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a diagram of an application environment of a thermal power station remote sensing detection method based on adaptive Fourier convolution in one embodiment;

[0039] Figure 2 A schematic diagram of a flow chart of a remote sensing detection method for a thermal power station based on adaptive Fourier convolution in one embodiment;

[0040] Figure 3 Schematic diagram of the structure of an adaptive Fourier convolution layer in one embodiment;

[0041] Figure 4 A schematic diagram of the architecture of each network layer in an encoder structure and a decoder mechanism in one embodiment;

[0042] Figure 5 A simplified flowchart of a remote sensing detection method for a thermal power station based on adaptive Fourier convolution in one embodiment;

[0043] Figure 6 Results of experiments comparing various network methods in one embodiment and a visual evaluation comparison chart;

[0044] Figure 7It is a structural block diagram of a thermal power station remote sensing detection system based on adaptive Fourier convolution in one embodiment;

[0045] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe asymmetric convolutional layers, but these asymmetric convolutional layers are not limited by these terms. These terms are only used to distinguish the first asymmetric convolutional layer from another asymmetric convolutional layer. For example, without departing from the scope of the present application, the first asymmetric convolutional layer may be referred to as the second asymmetric convolutional layer, and similarly, the second asymmetric convolutional layer may be referred to as the first asymmetric convolutional layer. Both the first asymmetric convolutional layer and the second asymmetric convolutional layer are asymmetric convolutional layers, but they are not the same asymmetric convolutional layer.

[0048] The thermal power station remote sensing detection method based on adaptive Fourier convolution provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can collect remote sensing images of the thermal power plant area through remote sensing technology, and perform data preprocessing operations on the remote sensing images to obtain processed remote sensing images; the computer device 110 can input the processed remote sensing images into the encoder structure, and process the processed remote sensing images through the adaptive Fourier convolution layer in the encoder structure to obtain feature maps; the computer device 110 can input the feature maps into the decoder structure, and reconstruct and identify the exhaust characteristics of the cooling tower in the feature maps through the decoder mechanism, and output classification results; the computer device 110 can determine the working state of the thermal power plant according to the classification results; wherein the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part; the decoder mechanism is provided with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part. Among them, the computer device 110 can be, but is not limited to, various personal computers, laptops, robots and other devices.

[0049] In one embodiment, Figure 2 As shown, a remote sensing detection method for a thermal power station based on adaptive Fourier convolution is provided, comprising the following steps:

[0050] Step 202, collect remote sensing images of the thermal power station area through remote sensing technology, and perform data preprocessing operations on the remote sensing images to obtain processed remote sensing images.

[0051] Among them, collecting remote sensing images of the thermal power station area through remote sensing technology is the first step to achieve thermal power station detection. In this embodiment, the computer device can use high-spectral, high-spatial-resolution and high-temporal-resolution remote sensing technology to obtain remote sensing images covering the thermal power station area.

[0052] Specifically, hyperspectral imaging technology is an important part of modern remote sensing technology, especially suitable for detailed monitoring of complex industrial areas such as thermal power plants. By capturing the distribution of electromagnetic spectrum reflected or emitted from the target object, the spectral information of the object is obtained. Compared with traditional RGB images, hyperspectral imaging can record hundreds of continuous bands, providing detailed spectral data from near-ultraviolet to infrared for each pixel. Among them, in the monitoring of thermal power plants, hyperspectral imaging can accurately identify different materials and states, such as exhaust from cooling towers and other industrial emissions. These emissions often have unique spectral characteristics, such as spectral absorption peaks of specific chemical components, which enables hyperspectral imaging to effectively distinguish and monitor the diffusion of different chemicals. In addition, by comparing spectral data at different time points, it is possible to effectively track environmental changes, monitor the spatiotemporal distribution of pollutants, and evaluate the effectiveness of implementing environmental protection measures.

[0053] High spatial resolution remote sensing images are essential for accurately identifying and monitoring ground targets. In the monitoring of thermal power plants, high spatial resolution helps to show in detail the characteristics of each structure, such as cooling towers, chimneys, the main body of the power plant and the surrounding environment, and can provide enough details so that even small facilities or minor changes inside the power plant can be accurately captured. In addition, high spatial resolution images are particularly important in disaster emergency management. For example, when an accident such as a fire or structural damage occurs at a power plant, high-resolution images can quickly provide damage assessments to help decision makers and rescue teams develop effective response strategies. For routine maintenance, such images can also help engineers identify potential structural problems, prevent accidents, and ensure the safe operation of power plants.

[0054] Remote sensing technology with high temporal resolution can frequently acquire images of the same area, which is particularly important for monitoring the dynamic changes of thermal power plants. It can ensure real-time or quasi-real-time data acquisition, so that monitoring personnel can promptly understand the operating status of thermal power plants, including the exhaust of cooling towers and other visible environmental impacts. By analyzing continuous time series images, the changing trend of exhaust can be monitored to determine whether the power plant is operating normally. In terms of environmental supervision, high temporal resolution images are equally important. They can help environmental protection agencies monitor the compliance of power plant emissions, promptly detect excessive emission incidents, and take corresponding penalties or rectification measures. In extreme weather or disaster situations, this monitoring is also the key to ensuring the safe operation and timely response of power plants.

[0055] In one embodiment, a remote sensing detection method for a thermal power plant based on adaptive Fourier convolution is provided, which may also include a process of performing data preprocessing operations, the specific process including: using a filter to remove noise in a remote sensing image to obtain a remote sensing image after noise removal; correcting the remote sensing image after noise removal through radiation correction to obtain a corrected remote sensing image; and performing geometric correction on the corrected remote sensing image to obtain a processed remote sensing image.

[0056] In order to ensure the quality of remote sensing images, it is necessary to perform noise removal, radiation correction, and geometric correction operations on the collected remote sensing images.

[0057] Specifically, in one embodiment, the computer device may use a filter to remove noise in the remote sensing image by adopting a median filter, a mean filter, or a convolution filter.

[0058] Computer equipment can use filters and other image processing techniques to remove noise from remote sensing images and improve image quality. In the processing of remote sensing images, noise removal is an important step to improve image quality. Noise removal can be performed using methods such as median filtering, mean filtering, and convolution filtering to retain image detail information. Among them, median filtering is a nonlinear image processing technology that effectively removes salt and pepper noise from images by replacing each pixel value with the median of the pixel values ​​in its neighborhood; mean filtering smoothes the image by calculating the average value in the pixel neighborhood, which is suitable for removing random noise; convolution filtering uses a predefined kernel to perform a convolution operation with the image to achieve the effect of denoising and sharpening the image.

[0059] In one embodiment, when performing radiation correction, the computer device can obtain the received radiation value through the sensor and determine the actual radiation value of the ground; compare the received radiation value with the actual radiation value of the ground, and correct the remote sensing image after noise removal based on the comparison result to obtain a corrected remote sensing image.

[0060] The purpose of radiometric correction is to eliminate image radiometric inconsistencies caused by sensor characteristics, atmospheric conditions, and other external factors. Radiometric correction usually requires real data of actual ground radiation as a reference, and corrects the image by comparing the difference between the radiation value received by the sensor and the actual radiation value. In this embodiment, relative radiometric correction focuses on the radiometric consistency within the image, ensuring that the comparison between different parts of the image is fair, which is particularly suitable for change monitoring and multi-temporal analysis.

[0061] Specifically, in this embodiment, geometric correction is a key step to ensure that the image accurately reflects the geographic location. By using ground control points (GCPs) and digital elevation models (DEM), image distortions caused by sensor angles, flight altitudes, and ground undulations can be corrected. Geometrically corrected images can accurately correspond to geographic coordinate systems, which is critical for applications such as map making, resource management, and environmental monitoring.

[0062] Step 204, input the processed remote sensing image into the encoder structure, and process the processed remote sensing image through the adaptive Fourier convolution layer in the encoder structure to obtain a feature map.

[0063] Among them, the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part.

[0064] Specifically, the remote sensing image after preprocessing will enter the encoder structure, and the encoder accepts input remote sensing image data, and the number of input channels is 3 (corresponding to three RGB channels).

[0065] In one embodiment, a thermal power station remote sensing detection method with adaptive Fourier convolution is provided, which may also include a process of processing a remote sensing image through an encoder structure, and the specific process includes: performing Fourier transform on the processed remote sensing image through a Fourier domain pooling layer to extract the spectral features in the processed remote sensing image; performing an asymmetric convolution operation on the processed remote sensing image through a first asymmetric convolution layer to obtain the directional features in the processed remote sensing image; fusing the spectral features and directional features through a feature fusion part, and processing through batch normalization and activation function to obtain a feature map.

[0066] After the encoder receives the input remote sensing image data, it can process the input image through multiple adaptive Fourier convolution layers (Adaptive Fourier Convolution, AFC). The structure of the adaptive Fourier convolution layer is as follows: Figure 3 That is to say, a plurality of adaptive Fourier convolution layers are arranged in the encoder structure, and the number of channels in each adaptive Fourier convolution layer increases successively.

[0067] Specifically, each AFC layer includes a Fourier domain pooling layer, an asymmetric convolution layer, and a feature fusion part. The Fourier domain pooling layer performs Fourier transform on the input image data, extracts spectral features through operations in the Fourier domain, and effectively removes noise and interference in the image; the asymmetric convolution layer applies asymmetric convolution operations to flexibly capture directional features in the image; the feature fusion part combines Fourier domain features with convolution features, and processes them through batch normalization and activation functions to improve the robustness of feature extraction.

[0068] The encoder includes multiple AFC layers, gradually increasing the number of channels to extract higher-level features, from 3 channels to 64 channels in the first layer, to 64 channels to 128 channels in the second layer, to 128 channels to 256 channels in the third layer, and finally to 256 channels to 512 channels in the fourth layer.

[0069] In one embodiment, each AFC layer can be followed by a meta-attention mechanism module, which weights the features through the attention mechanism, emphasizes important features, suppresses unimportant features, and further improves the accuracy of feature extraction. In addition, adaptive multi-scale fusion technology is applied to multiple levels of the encoder, including convolution operations of different scales (e.g., 1x1 convolution, 3x3 convolution, 5x5 convolution, 7x7 convolution) to capture information of different scales, fuse multi-scale features, and generate richer and more comprehensive image features. The last layer of the encoder outputs a high-level feature map after the above processing, which provides input for the subsequent decoder or classifier.

[0070] Step 206, input the feature map into the decoder structure, reconstruct and identify the exhaust characteristics of the cooling tower in the feature map through the decoder mechanism, and output the classification result.

[0071] Among them, the decoder mechanism is equipped with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part. The decoder mechanism mainly reconstructs and identifies key information in the image from the high-level feature map extracted by the encoder, such as the cooling tower and its exhaust characteristics. Specifically, for high-precision remote sensing image detection tasks, after receiving the high-level feature map from the encoder mechanism, the decoder mechanism undergoes a series of key steps to accurately reconstruct and identify key information in the image. The design structure of the decoder mechanism makes full use of technologies such as adaptive multi-scale fusion, meta-attention mechanism, asymmetric convolution and patch expansion, providing reliable data support for the operation status monitoring of thermal power plants.

[0072] In one embodiment, a remote sensing detection method for a thermal power plant based on adaptive Fourier convolution is provided, which may also include a process of feature reconstruction and recognition, and the specific process includes: applying adaptive multi-scale fusion technology to the feature map through an adaptive multi-scale fusion layer to capture multi-scale features, and fusing the multi-scale features; reconstructing and recognizing the fused multi-scale features through a second asymmetric convolution layer and a patch expansion part to obtain a classification result.

[0073] Among them, the adaptive multi-scale fusion layer applies adaptive multi-scale fusion technology on multiple levels of the decoder, combining convolution operations of different scales, such as 1x1, 3x3, 5x5, and 7x7 convolutions, to capture multi-scale information and fuse these multi-scale features to generate more comprehensive image features, gradually reducing the number of channels to match the size of feature maps after upsampling at different levels, ensuring the proper transmission and processing of information.

[0074] The application of the second asymmetric convolutional layer adds flexibility to the decoder, enabling it to better capture directional features in the image, thereby improving the effect of feature reconstruction. These second asymmetric convolutional layers are alternated with adaptive multi-scale fusion layers to restore the spatial resolution of the image layer by layer. Each second asymmetric convolutional layer is followed by a patch expansion layer, which gradually restores the spatial resolution of the image through upsampling operations until it is finally restored to the original image size. This layer-by-layer processing helps ensure the accuracy and completeness of image restoration.

[0075] The output layer is the last layer of the decoder, which is a convolutional layer (Conv 1x1), which converts the processed feature map into the final classification result. The number of input channels is set to 64, and the number of output channels is set to the required number of classification channels to ensure that the final output meets the expected classification requirements.

[0076] Specifically, in one embodiment, the patch expansion part is arranged after the second asymmetric convolution layer, and the spatial resolution of the image is gradually restored through upsampling operations to complete feature image reconstruction.

[0077] In one embodiment, a meta-attention mechanism module is provided in the decoder mechanism for feature weighting. The meta-attention mechanism module is introduced, following each layer, and features are weighted through the attention mechanism, emphasizing important features while suppressing unimportant features, further improving the accuracy of feature reconstruction, and helping to better focus on key information during feature reconstruction, thereby improving the performance and accuracy of the model.

[0078] In this embodiment, the structural design of the entire decoder enables it to efficiently and accurately reconstruct and identify key information in images, such as cooling towers and exhaust features in thermal power plants. The decoder combines multiple technologies such as adaptive multi-scale fusion, asymmetric convolution, meta-attention mechanism, and patch expansion to effectively improve the accuracy and efficiency of feature reconstruction; it is not only suitable for image processing in simple environments, but also more suitable for high-precision remote sensing image detection tasks in complex environments, providing more reliable data support for monitoring and analysis.

[0079] Step 208, determining the working status of the thermal power station according to the classification result.

[0080] In one embodiment, in a thermal power plant remote sensing detection method based on adaptive Fourier convolution provided in the present application, the encoder structure and each network layer architecture in the decoder mechanism are as follows: Figure 4 As shown in FIG. 1 , each layer in the encoder structure includes an input layer, an adaptive Fourier convolution AFC layer 1, a meta-attention mechanism module, an AFC layer 2, a meta-attention mechanism module, an AFC layer 3, a meta-attention mechanism module, and an AFC layer 4. The number of input channels, the number of output channels, the operation description, and the technical features of each layer are shown in FIG. Figure 4 The decoder structure includes input layer, adaptive fusion layer, asymmetric convolution layer, patch expansion layer, adaptive fusion layer, asymmetric convolution layer, and output layer. The number of input channels, the number of output channels, the operation description, and the technical features of each layer are shown in Figure 1. Figure 4 shown.

[0081] like Figure 5 As shown, in one embodiment, a remote sensing detection method for a thermal power plant based on adaptive Fourier convolution is provided. By designing a technology based on an adaptive Fourier convolution layer, including a Fourier domain pooling layer, an asymmetric convolution, a meta-attention mechanism, and an adaptive multi-scale fusion, it is possible to efficiently learn and extract fine-grained spectral features in an image, effectively remove noise and interference, flexibly capture directional features and multi-scale information in an image, and improve the accuracy and efficiency of feature extraction.

[0082] Specifically, Figure 5As shown, the process of a remote sensing detection method for a thermal power station based on adaptive Fourier convolution may include: data acquisition, data preprocessing, encoder processing, decoder processing, and data output. Specifically, in this embodiment, the computer equipment can obtain image data of the thermal power station through remote sensing technology and analyze the exhaust conditions of the cooling tower; the adaptive Fourier convolution technology can accurately identify the cooling tower and its exhaust characteristics in the image, so as to judge the working state of the thermal power station; combined with the multi-scale feature extraction method, the high-frequency information of the neural network is first learned, and then the classification and judgment are performed through the deep learning model (such as LSTM), wherein the multi-scale feature extraction helps to maintain high-precision detection in complex environments and improve the ability to identify the exhaust conditions of the cooling tower.

[0083] In this embodiment, a remote sensing detection method for a thermal power station based on adaptive Fourier convolution is provided. By adopting adaptive Fourier convolution (AFC) network technology, the ability to accurately detect remote sensing images of thermal power stations using intelligent inspection technology in dynamic and complex environments is significantly improved; while ensuring that the strict requirements of industrial applications are met, it also shows excellent compatibility and scalability, and is particularly suitable for thermal power stations, which have extremely high requirements for detection accuracy and environmental adaptability. The core advantage of the adaptive Fourier convolution network is that it can comprehensively utilize spatial and frequency domain information. By integrating these two types of information in each layer of the network, the recognition accuracy of image features is greatly improved; in the remote sensing image detection task of the thermal power station, this innovation enables the AFC network to effectively identify and process visual interference caused by environmental factors such as smoke and dust, thereby ensuring the accuracy and reliability of the detection results. In a dynamic and complex environment, such as the daily operation scene of a thermal power station, the AFC network shows excellent environmental adaptability. Whether it is the visual changes caused by changes in illumination, seasonal changes, or equipment aging, the AFC network can adjust the processing strategy through its adaptive mechanism to maintain a high level of detection performance.

[0084] In one embodiment, in order to verify the effect of a thermal power plant remote sensing detection method based on adaptive Fourier convolution, the adaptive Fourier convolution (AFC) network used in the thermal power plant remote sensing detection method based on adaptive Fourier convolution in this application is experimentally compared with other network detection methods such as U-Net, SegNet, DeepLabv3+, and FCN. The comparison results and visual evaluation are as follows: Figure 6 As shown. Figure 6It can be seen that the adaptive Fourier convolution (AFC) network outperforms traditional segmentation network models such as U-Net, SegNet, DeepLabv3+, and FCN in terms of overall accuracy and recognition of cooling tower features. The AFC network especially demonstrates its advantages in visual evaluation, and can more clearly reconstruct the structural details and exhaust morphology of the cooling tower, thanks to its efficient ability to process frequency domain features and spatial details. These results support the strategy of using the AFC network when performing high-precision remote sensing image processing in complex environments.

[0085] In this experiment, U-Net is a network structure widely used in medical image segmentation. Its design advantage is that it can capture rich contextual information by using a large number of skip connections. However, in this experiment, the overall accuracy of U-Net is 88%, the precision and recall rate in the recognition of cooling tower features are 90% and 89% respectively, and the F1 score is 89.5%. Although these values ​​are good, they are slightly insufficient compared with AFC. The visual evaluation mentioned that U-Net is fuzzy in terms of structural details, which may be because it is not as fine as the specially designed network in capturing high-frequency details when processing non-medical images.

[0086] SegNet is designed primarily to solve road and scene parsing problems, and it restores image details by mapping low-resolution features to high-resolution outputs through an effective decoder. In this experiment, SegNet achieved an overall accuracy of 85%, with slightly lower precision and recall of 88% and 87% for the recognition of cooling tower features, and an F1 score of 87.5%. These results suggest that SegNet may not be as effective as other models in capturing details in complex industrial environments, which is also confirmed by its performance in visual evaluation, where detail reconstruction is poor when dealing with exhaust features of complex shapes.

[0087] DeepLabv3+ is an advanced segmentation network that combines dilated convolution and encoder-decoder paths, designed to improve segmentation boundaries and enhance the model's ability to understand multi-scale information. In this experiment, DeepLabv3+ achieved an overall accuracy of 90%, with a recognition accuracy and recall of 92% and 91% for cooling tower features, and an F1 score of 91.5%, indicating that DeepLabv3+ has better performance when processing images with complex backgrounds, but is still inferior to AFC in the precise expression of details.

[0088] FCN (Fully Convolutional Network) is one of the pioneering works in the field of image segmentation. It first proposed end-to-end segmentation using convolutional networks. FCN performed relatively poorly in this experiment, with an overall accuracy of 86%, a precision and recall of 87% and 86% for the recognition of cooling tower features, and an F1 score of 86.5%. Although FCN is relatively accurate in recognizing the overall shape, it is not good at restoring image details, especially in complex environments.

[0089] It should be understood that, although the steps in the above-mentioned flowcharts are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flowcharts may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0090] In one embodiment, Figure 7 As shown, a remote sensing detection system for a thermal power station based on adaptive Fourier convolution is provided, comprising: an image acquisition and processing module 710, an encoder processing module 720, a decoder processing module 730 and a state determination module 740, wherein:

[0091] The image acquisition and processing module 710 is used to acquire remote sensing images of the thermal power station area through remote sensing technology, and perform data preprocessing operations on the remote sensing images to obtain processed remote sensing images;

[0092] The encoder processing module 720 is used to input the processed remote sensing image into the encoder structure, and process the processed remote sensing image through the adaptive Fourier convolution layer in the encoder structure to obtain a feature map;

[0093] A decoder processing module 730 is used to input the feature map into the decoder structure, reconstruct and identify the exhaust characteristics of the cooling tower in the feature map through the decoder mechanism, and output the classification result;

[0094] A state determination module 740, used to determine the working state of the thermal power station according to the classification result;

[0095] Among them, the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part; the decoder mechanism is provided with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part.

[0096] In one embodiment, the image acquisition and processing module 710 is also used to use a filter to remove noise from the remote sensing image to obtain a remote sensing image after noise removal; correct the remote sensing image after noise removal through radiation correction to obtain a corrected remote sensing image; and perform geometric correction on the corrected remote sensing image to obtain a processed remote sensing image.

[0097] In one embodiment, the image acquisition and processing module 710 is further used to use a filter to remove noise in the remote sensing image by means of median filtering, mean filtering or convolution filtering.

[0098] In one embodiment, the image acquisition and processing module 710 is also used to obtain the received radiation value through the sensor and determine the actual radiation value of the ground; compare the received radiation value with the actual radiation value of the ground, and correct the remote sensing image after noise removal based on the comparison result to obtain a corrected remote sensing image.

[0099] In one embodiment, the encoder processing module 720 is also used to perform Fourier transform on the processed remote sensing image through a Fourier domain pooling layer to extract spectral features in the processed remote sensing image; perform an asymmetric convolution operation on the processed remote sensing image through a first asymmetric convolution layer to obtain directional features in the processed remote sensing image; fuse the spectral features and directional features through a feature fusion part, and process them through batch normalization and activation functions to obtain a feature map.

[0100] In one embodiment, a plurality of adaptive Fourier convolution layers are provided in the encoder structure, and the number of channels in each adaptive Fourier convolution layer increases sequentially.

[0101] In one embodiment, the decoder processing module 730 is also used to apply adaptive multi-scale fusion technology to the feature map through an adaptive multi-scale fusion layer to capture multi-scale features, and fuse the multi-scale features; the fused multi-scale features are reconstructed and identified through a second asymmetric convolution layer and a patch expansion part to obtain a classification result.

[0102] In one embodiment, the adaptive multi-scale fusion layer and the second asymmetric convolution layer alternately perform feature processing on the feature map; the patch expansion part is arranged after the second asymmetric convolution layer, and the spatial resolution of the image is gradually restored through upsampling operations to complete feature image reconstruction.

[0103] In one embodiment, a meta-attention mechanism module is provided after the adaptive Fourier convolution layer and in the decoder mechanism for performing feature weighting.

[0104] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a remote sensing detection method for a thermal power station based on adaptive Fourier convolution is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0105] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0106] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of a remote sensing detection method for a thermal power station based on adaptive Fourier convolution are implemented.

[0107] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a remote sensing detection method for a thermal power station based on adaptive Fourier convolution are implemented.

[0108] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0109] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A remote sensing detection method for a thermal power station based on adaptive Fourier convolution, characterized in that: The method comprises: Collecting remote sensing images of the thermal power station area through remote sensing technology, and performing data preprocessing operations on the remote sensing images to obtain processed remote sensing images; Inputting the processed remote sensing image into an encoder structure, and processing the processed remote sensing image through an adaptive Fourier convolution layer in the encoder structure to obtain a feature map; Inputting the feature map into a decoder structure, reconstructing and identifying the exhaust characteristics of the cooling tower in the feature map through the decoder mechanism, and outputting a classification result; determining the working state of the thermal power station according to the classification result; Among them, the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part; the decoder mechanism is provided with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part.

2. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 1 is characterized in that: The performing of a data preprocessing operation on the remote sensing image to obtain a processed remote sensing image includes: Using a filter to remove noise from the remote sensing image to obtain a remote sensing image after noise removal; The remote sensing image after noise removal is corrected by radiation correction to obtain a corrected remote sensing image; The corrected remote sensing image is geometrically corrected to obtain a processed remote sensing image.

3. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 2 is characterized in that: The step of removing noise from the remote sensing image by using a filter comprises: The filter is used to remove the noise in the remote sensing image by means of median filtering, mean filtering or convolution filtering.

4. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 2 is characterized in that: The step of correcting the remote sensing image after noise removal by radiation correction to obtain a corrected remote sensing image includes: Obtain the received radiation value through the sensor and determine the actual radiation value on the ground; The received radiation value is compared with the actual ground radiation value, and based on the comparison result, the remote sensing image after noise removal is corrected to obtain a corrected remote sensing image.

5. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 1, characterized in that: The processed remote sensing image is processed by the adaptive Fourier convolution layer in the encoder structure to obtain a feature map, including: Performing Fourier transform on the processed remote sensing image through the Fourier domain pooling layer to extract spectral features in the processed remote sensing image; Performing an asymmetric convolution operation on the processed remote sensing image through the first asymmetric convolution layer to obtain directional features in the processed remote sensing image; The spectral features and directional features are fused through the feature fusion part, and processed through batch normalization and activation function to obtain a feature map.

6. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 5, characterized in that: A plurality of the adaptive Fourier convolution layers are arranged in the encoder structure, and the number of channels in each of the adaptive Fourier convolution layers increases sequentially.

7. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 1, characterized in that: The reconstructing and identifying the exhaust characteristics of the cooling tower in the characteristic graph by the decoder mechanism and outputting the classification result comprises: Applying an adaptive multi-scale fusion technology to the feature map through the adaptive multi-scale fusion layer to capture multi-scale features, and fusing the multi-scale features; The fused multi-scale features are passed through the second asymmetric convolutional layer and the patch expansion part to perform feature reconstruction and recognition to obtain a classification result.

8. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 7, characterized in that: The adaptive multi-scale fusion layer and the second asymmetric convolution layer alternately perform feature processing on the feature map; The patch expansion part is arranged after the second asymmetric convolution layer, and gradually restores the spatial resolution of the image through upsampling operation to complete the feature image reconstruction.

9. The method for remote sensing detection of a thermal power station based on adaptive Fourier convolution according to claim 1, characterized in that: A meta-attention mechanism module is provided behind the adaptive Fourier convolution layer and in the decoder mechanism for performing feature weighting.

10. A remote sensing detection system for a thermal power station based on adaptive Fourier convolution, characterized in that: The system comprises: An image acquisition and processing module is used to acquire remote sensing images of the thermal power station area through remote sensing technology, and perform data preprocessing operations on the remote sensing images to obtain processed remote sensing images; An encoder processing module, used for inputting the processed remote sensing image into an encoder structure, and processing the processed remote sensing image through an adaptive Fourier convolution layer in the encoder structure to obtain a feature map; A decoder processing module, used for inputting the feature map into a decoder structure, reconstructing and identifying the exhaust characteristics of the cooling tower in the feature map through the decoder mechanism, and outputting a classification result; A state determination module, used to determine the working state of the thermal power station according to the classification result; Among them, the adaptive Fourier convolution layer includes a Fourier domain pooling layer, a first asymmetric convolution layer, and a feature fusion part; the decoder mechanism is provided with an adaptive multi-scale fusion layer, a second asymmetric convolution layer, and a patch expansion part.

Images