Image defogging enhancement method and system for fully mechanized coal mining face in dust fog environment

Abstract

The invention discloses an image defogging enhancement method and system for a fully mechanized coal mining face in a dust fog environment, and the method comprises the steps: obtaining a to-be-processed foggy image, carrying out the encoding operation of the foggy image through an encoder, and obtaining an encoding feature from the foggy image; inputting the obtained coding features into a bottleneck layer integrated by a DM module to realize feature coupling of the coding features in a space domain and a frequency domain to obtain fusion features; and carrying out decoding processing on the obtained fusion features through a decoder, and outputting to obtain a target defogged image. The system comprises a camera and an image defogging network model module based on double-domain coupling. The non-uniform haze feature representation and global feature perception are enhanced through the deformable convolution and amplitude phase guidance strategy, the image feature extraction definition of the model is improved, the details and structure of the image are restored, and the defogging effect and generalization and generalization ability of the model are improved.

Description

Technical Field

[0001] The invention relates to an image defogging and enhancement method and system in a dusty and foggy environment of a fully-mechanized mining working face in a coal mine, belonging to the technical field of image processing. Background Art

[0002] Image dehazing is a key task in the field of computer vision, which aims to restore clear and high-quality images that have been degraded by haze. In the dusty and foggy environment of coal mine fully mechanized working faces, the working face involves mechanized operations such as coal mining, coal loading, coal transportation, support, goaf treatment and tunnel excavation, which will generate a large amount of coal dust and dust in this process. In addition, in order to control the concentration of gas and dust, water spraying and spraying are usually used to suppress dust, which further aggravates the concentration of dust and fog. Traditional image dehazing algorithms usually rely on manually designed image priors. Such methods obtain prior knowledge by statistically analyzing the relationship between blurred images and clear images, but their specific assumptions limit their generalization ability when dealing with complex blurred images and unseen scenes.

[0003] With the advancement of deep convolutional neural networks, significant progress has been made in the field of image dehazing. Such methods have achieved excellent results in peak signal-to-noise ratio and structural similarity index on SOTS-Indoor through supervised training on large datasets. Recently, with the introduction of Transformer in the field of image dehazing, the ability to effectively capture global information and long-range dependencies in images has been enhanced, improving the feature representation of images and achieving state-of-the-art performance. Nevertheless, the quadratic complexity of self-attention remains a huge computational challenge. In addition, texture features may show strong correlation or extension in specific directions, and the degradation caused by haze is almost completely reflected in the amplitude component, while the difference in the phase component between hazy and clear images is small.

[0004] In summary, the problems of existing image dehazing technology are mainly manifested as follows: fixed-size convolution is difficult to achieve inter-channel interaction and cannot effectively extract multi-scale information, making it difficult to accurately capture the haze distribution and reconstruct clear texture details in dense haze areas; it mainly focuses on extracting spatial features in the spatial domain, and the image feature extraction effect is poor. Summary of the invention

[0005] The purpose of the invention is to provide an image dehazing and enhancement method and system in a dusty and foggy environment of a fully-mechanized mining face in a coal mine. The method and system can improve the clarity of image feature extraction, restore the details and structure of the image, enhance the generalization ability of the model, and ensure the safe mining of the coal mine.

[0006] In order to achieve the above object, the present invention provides an image defogging and enhancement method in a dusty and foggy environment of a fully-mechanized coal mining working face, comprising the following steps:

[0007] S1. Obtain a foggy image to be processed, and perform an encoding operation on the foggy image through an encoder to obtain encoding features from the foggy image;

[0008] S2, input the obtained coding features into the bottleneck layer integrated by the DM module to achieve the coupling of the coding features in the spatial domain and the frequency domain to obtain the fusion features;

[0009] S3. The obtained fusion features are decoded by a decoder and the target defogging image is output.

[0010] Furthermore, the specific process of S1 is as follows:

[0011] S1.1. Use the RESIDE dataset to obtain foggy images and clear images respectively, randomly crop the foggy images to 256×256, and horizontally flip the foggy images with a probability of 50% to train the defogging network;

[0012] S1.2, the defogging network includes an encoder, a bottleneck layer and a decoder. The encoder performs three downsampling operations to double the feature channel, and the width and height become half of the original. Each downsampling operation consists of four ResBlock blocks, and each ResBlock block consists of two 3×3 convolutions and an intermediate GELU activation function repeated four times; the encoding features corresponding to the foggy image features are gradually extracted through the ResBlock blocks of the encoder. Each ResBlock block has a stronger ability to extract encoding features than the previous ResBlock block. The multi-level structure of the encoder can capture and process image degradation features at different scales, thereby improving the robustness of defogging.

[0013] Furthermore, the DM module in S2 includes a convolutional layer, a DCM module, an APGM module and an SRU module; the fusion process of the fusion feature is:

[0014] S2.1, through the deformable convolution operator DCM, the spatial context information is effectively aggregated and the texture details of the heavy haze area are reconstructed. For the input features DC-h and DC-w operations are applied to extract context information from the height and width directions respectively. The formula is as follows:

[0015]

[0016] In the formula, ω i represents the parameters of the deformable convolution, o c for the width and height dimensions in channel c i The learnable offset on ;

[0017] S2.2. Calculate the average value of all pixels in each channel, apply 1×1 point convolution to double the number of channels, and perform softmax operation to obtain the weight ω:

[0018]

[0019] ω=softmax(conv(X avg ));

[0020] Where, X avg Represents X DC-h With X DC-w The average value of all pixels in each channel, conv represents 1×1 convolution, ω represents X DC-h With X DC-w weights array;

[0021] S2.3, weight ω and X DC-h and X DC-w Multiply and convolve to produce the final output:

[0022] X DCM =ω[0]*X DC-h +ω[1]*X DC-w ;

[0023] In the formula, ω[0] and ω[1] represent X DC-h and X DC-h The weight of X DCM The output result of the DCM module. Through the DCM module, the model effectively enhances the ability to capture haze distribution, which is conducive to extracting clear feature information.

[0024] S2.4, through the amplitude phase guidance module APGM, the prior knowledge of amplitude and phase is used to guide the extraction of global features:

[0025] S2.4.1. For input feature map Apply fast Fourier transform FFT to convert the input features from the spatial domain to the frequency domain, and obtain the amplitude and phase information as follows:

[0026]

[0027] phase = atan(B,A);

[0028] In the formula, is a fast Fourier transformer, A and B represent the real part and the imaginary part respectively, and atan() is the inverse tangent function;

[0029] S2.4.2. Perform low-frequency filtering on the amplitude and high-frequency filtering on the phase. Apply a residual block ResBlock consisting of two convolutional layers with a LeakyReLU activation function in the middle to the amplitude and phase components, and then combine the original and modulated amplitude and phase information through an inverse Fourier transform to produce a feature map containing phase and amplitude modulation:

[0030] amb'=conv 1*1 (LeakyReLu(conv 1*1 (amb)));

[0031] phase'=conv 1*1 (LeakyReLu(conv 1*1 (phase)));

[0032]

[0033] In the formula, e i represents a complex exponential, It is the inverse operation of the fast Fourier transform;

[0034] S2.4.3. Connect the modulated feature maps along the channel dimension, and then perform 1×1 point convolution to reduce the number of channels by half to obtain the final output:

[0035] X APGM =conv 1*1 (concat(X amb ,X angel ));

[0036] Where, X APGM Represents the output result of the APGM module; this operation realizes the mutual guidance between amplitude and phase, and enhances the ability of the model to capture global information more effectively;

[0037] S2.5, Fusion of spatial and frequency domain features: The input features are processed in parallel through 1×1 point convolution, DCM module, and APGM module, and then the output features are combined using addition operation to generate reconstruction information combining spatial and frequency domains:

[0038] S2.5.1. The obtained output features are combined by addition operation to generate reconstruction information combining spatial domain and frequency domain, which is expressed as:

[0039] F = conv 1×1 (X)++X DCM (X)++X APGM (X);

[0040] In the formula, conv 1×1represents the output feature of the convolution operation with a convolution kernel of 1×1, X DCM and X APGM Indicates the output characteristics of the DCM module and APGM module;

[0041] S2.5.2. The spatial domain and frequency domain are coupled through the spatial reconstruction module SRU. For the input features First, we use BatchNorm to normalize the weight map, and then use the Sigmoid function to map the weight map to the range (0,1). Then, we perform a soft threshold operation to set the weights above a certain threshold to 1 to obtain information weights; the weights below the threshold are set to 0 to form non-information weights. These weights are multiplied by the input feature F to generate two weighted features, namely, information-rich features and less information-rich features, as shown below:

[0042] F w =σ(BatchNorm(F));

[0043]

[0044] Where σ is the sigmoid function, F w represents the weight of input feature F, F 1 and F 2 respectively, information-rich features and less-informative features;

[0045] S2.5.3. In order to further refine the information content and optimize the space, a reconstruction operation is performed to fuse the two weighted features and enhance the information flow between them. Finally, the reconstructed features are connected to generate the final feature map. It is expressed as:

[0046]

[0047] Furthermore, the method of S3 is as follows: the decoder performs three upsampling operations to reduce the feature channel to half of the original, and the width and height are doubled; each upsampling operation consists of four ResBlock blocks, and each ResBlock block consists of two 3×3 convolutions and an intermediate GELU activation function repeated four times; after the last ResBlock block, the feature is mapped to the RGB space through a 3×3 convolution to output the dehazed image; the specific process includes:

[0048] S3.1. The fusion feature is recorded as The encoder is recorded as Decoder, and the dehazed image is obtained after being processed by the decoder It is expressed as:

[0049]

[0050] S3.2. To ensure the consistency between the dehazed image and the clear image at the pixel level and in the frequency domain, L1 loss, frequency domain L1 loss, amplitude contrast loss, and phase contrast loss are used as the main optimization objectives. The loss function is as follows:

[0051]

[0052]

[0053] In the formula, represents the dehazed image, Y represents the clear image, They represent L1 loss, frequency domain L1 loss, amplitude contrast loss, and phase contrast loss respectively. A() and P() represent amplitude and phase information respectively. The overall network loss is expressed as follows:

[0054]

[0055] In the formula, λ 1 ,λ 2 ,λ 3 They are empirically set to 0.1, 0.9, and 0.2, respectively, to balance the dual-domain training.

[0056] An image defogging and enhancement system in a dusty and foggy environment of a fully-mechanized coal mining working face, comprising a camera and a dual-domain coupling-based image defogging network model module;

[0057] The camera is used to collect dust and fog environment pictures of the fully-mechanized mining working face in the coal mine, save them to a local path, and write the foggy image path into a database for image defogging;

[0058] The dual-domain coupled image defogging network model module is used to defog a foggy image to obtain a restored image; an optimized defogging model is obtained by iteratively training the dual-domain coupled image defogging network on the RESIDE-Outdoor dataset.

[0059] The dual-domain coupled image dehazing network is trained as follows: the network is trained for 30 iterations on an RTX 3090, and the training BatchSize is set to 8; in order to achieve better model convergence, the present invention uses the Adam optimizer and sets the parameter β 1 and β 2 Set to 0.9, 0.99; In order to improve the model training efficiency and performance, the present invention adopts the cosine annealing decay strategy to reduce the learning rate from 2×10 -4 Gradually reduce to 1×10 -6 ;

[0060] The trained network model is loaded, the foggy image path is obtained from the database, the foggy image is read, and the defogged image is output after one forward propagation, thus completing the construction of the image defogging enhancement system in the dusty and foggy environment of the coal mine fully mechanized mining working face.

[0061] The present invention enhances the representation of non-uniform haze features and global feature perception through deformable convolution and amplitude-phase guidance strategies. The deformable convolution operator DCM based on the spatial non-uniform strategy of channel interaction effectively aggregates spatial context information to process non-uniform haze distribution and reconstruct fine texture details in heavy haze areas by capturing features along the image height and width channels; the amplitude-phase guidance module APGM is used to implement low-pass filtering on the amplitude component and high-pass filtering on the phase component to effectively extract global features; by combining the DCM module and the APGM module, the DM module is inserted into the encoder-decoder bottleneck layer to overcome the obstacles faced in achieving spatial and frequency domain fusion, and finally the encoder-decoder architecture is supervised and trained to output a defogging image. The present invention improves the clarity of the model for image feature extraction, restores the details and structure of the image, improves the defogging effect and generalization ability of the model, and ensures the safe mining of coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a work flow chart of the method of the present invention;

[0063] Figure 2 It is a DM module diagram of the present invention;

[0064] Figure 3 It is a DCM module diagram of the present invention;

[0065] Figure 4 It is the APGM module diagram of the present invention;

[0066] Figure 5 It is a SRU module diagram of the present invention;

[0067] Figure 6 It is a schematic diagram of the working principle of the system of the present invention;

[0068] Figure 7 This is a comparison chart of the defogging effects of the embodiment of the present invention and the existing algorithm on the SOTS-Outdoor dataset;

[0069] Figure 8 It is a comparison diagram of the defogging effect of the embodiments of the present invention in a fully mechanized mining working face environment in a coal mine. DETAILED DESCRIPTION

[0070] The present invention will be further described below in conjunction with the accompanying drawings.

[0071] like Figure 1As shown, a method for image defogging and enhancement in a dusty and foggy environment of a fully-mechanized coal mining working face includes the following steps:

[0072] S1. Obtain a foggy image to be processed, and perform an encoding operation on the foggy image through an encoder to obtain encoding features from the foggy image;

[0073] S2, input the obtained coding features into the bottleneck layer integrated by the DM module to achieve the coupling of the coding features in the spatial domain and the frequency domain to obtain the fusion features;

[0074] S3. The obtained fusion features are decoded by a decoder and the target defogging image is output.

[0075] As a preferred implementation, the specific process of S1 is:

[0076] S1.1. Use the RESIDE dataset to obtain foggy images and clear images, randomly crop the foggy images to 256×256, and horizontally flip the foggy images with a probability of 50% to train the defogging network.

[0077] S1.2, the defogging network includes an encoder, a bottleneck layer and a decoder. The encoder performs three downsampling operations to double the feature channel, and the width and height become half of the original. Each downsampling operation consists of four ResBlock blocks, and each ResBlock block consists of two 3×3 convolutions and an intermediate GELU activation function repeated four times; the encoding features corresponding to the foggy image features are gradually extracted through the ResBlock blocks of the encoder. Each ResBlock block has a stronger ability to extract encoding features than the previous ResBlock block. The multi-level structure of the encoder can capture and process image degradation features at different scales, thereby improving the robustness of defogging.

[0078] Furthermore, if Figure 2 As shown, the DM module in S2 includes a convolutional layer, a DCM module, an APGM module and an SRU module; the fusion process of the fusion features is:

[0079] S2.1, such as Figure 3 As shown in the figure, the deformable convolution operator DCM effectively aggregates spatial context information and reconstructs the texture details of the heavy haze area. Apply DC-h and DC-w operations to extract context information from height and width respectively. The formula is as follows:

[0080]

[0081] In the formula, ω i represents the parameters of the deformable convolution, o cfor the width and height dimensions in channel c i The learnable offset on ;

[0082] S2.2. Calculate the average value of all pixels in each channel, apply 1×1 point convolution to double the number of channels, and perform softmax operation to obtain the weight ω:

[0083]

[0084] Where, X avg Represents X DC-h With X DC-w The average value of all pixels in each channel, conv represents 1×1 convolution, ω represents X DC-h With X DC-w weights array;

[0085] S2.3, weight ω and X DC-h and X DC-w Multiply and convolve to produce the final output:

[0086] X DCM =ω[0]*X DC-h +ω[1]*X DC-w ;

[0087] In the formula, ω[0] and ω[1] represent X DC-h and X DC-h The weight of X DCM The output result of the DCM module. Through the DCM module, the model effectively enhances the ability to capture haze distribution, which is conducive to extracting clear feature information.

[0088] S2.4, such as Figure 4 As shown in the figure, the amplitude phase guidance module APGM uses the prior knowledge of amplitude and phase to guide the extraction of global features:

[0089] S2.4.1. For input feature map Apply fast Fourier transform FFT to convert the input features from the spatial domain to the frequency domain, and obtain the amplitude and phase information as follows:

[0090]

[0091]

[0092] phase = atan(B,A);

[0093] In the formula, is a fast Fourier transformer, A and B represent the real part and the imaginary part respectively, and atan() is the inverse tangent function;

[0094] S2.4.2. Perform low-frequency filtering on the amplitude and high-frequency filtering on the phase. Apply a residual block ResBlock consisting of two convolutional layers with a LeakyReLU activation function in the middle to the amplitude and phase components, and then combine the original and modulated amplitude and phase information through an inverse Fourier transform to produce a feature map containing phase and amplitude modulation:

[0095] amb'=conv 1*1 (LeakyReLu(conv 1*1 (amb)));

[0096] phase'=conv 1*1 (LeakyReLu(conv 1*1 (phase)));

[0097]

[0098] In the formula, e i represents a complex exponential, It is the inverse operation of the fast Fourier transform;

[0099] S2.4.3. Connect the modulated feature maps along the channel dimension, and then perform 1×1 point convolution to reduce the number of channels by half to obtain the final output:

[0100] X APGM =conv 1*1 (concat(X amb ,X angel ));

[0101] Where, X APGM Represents the output result of the APGM module; this operation realizes the mutual guidance between amplitude and phase, and enhances the ability of the model to capture global information more effectively;

[0102] S2.5, Fusion of spatial and frequency domain features: The input features are processed in parallel through 1×1 point convolution, DCM module, and APGM module, and then the output features are combined using addition operation to generate reconstruction information combining spatial and frequency domains:

[0103] S2.5.1. The obtained output features are combined by addition operation to generate reconstruction information combining spatial domain and frequency domain, which is expressed as:

[0104] F=DW 1×1 (X)++X DCM (X)++X APGM (X);

[0105] In the formula, conv 1×1represents the output feature of the convolution operation with a convolution kernel of 1×1, X DCM and X APGM Indicates the output characteristics of the DCM module and APGM module;

[0106] S2.5.2, if Figure 5 As shown, the spatial domain and frequency domain are coupled through the spatial reconstruction module SRU. First, we use BatchNorm to normalize the weight map, and then use the Sigmoid function to map the weight map to the range (0,1). Then, we perform a soft threshold operation to set the weights above a certain threshold to 1 to obtain information weights; the weights below the threshold are set to 0 to form non-information weights. These weights are multiplied by the input feature F to generate two weighted features, namely, information-rich features and less information-rich features, as shown below:

[0107] F w =σ(BatchNorm(F));

[0108]

[0109] Where σ is the sigmoid function, F w represents the weight of input feature F, F 1 and F 2 respectively, information-rich features and less-informative features;

[0110] S2.5.3. In order to further refine the information content and optimize the space, a reconstruction operation is performed to fuse the two weighted features and enhance the information flow between them. Finally, the reconstructed features are connected to generate the final feature map. It is expressed as:

[0111]

[0112] Furthermore, the method of S3 is as follows: the decoder performs three upsampling operations to reduce the feature channel to half of the original, and the width and height are doubled; each upsampling operation consists of four ResBlock blocks, and each ResBlock block consists of two 3×3 convolutions and an intermediate GELU activation function repeated four times; after the last ResBlock block, the feature is mapped to the RGB space through a 3×3 convolution to output the dehazed image; the specific process includes:

[0113] S3.1. The fusion feature is recorded as The encoder is recorded as Decoder, and the dehazed image is obtained after being processed by the decoder It is expressed as:

[0114]

[0115] S3.2. To ensure the consistency between the dehazed image and the clear image at the pixel level and in the frequency domain, L1 loss, frequency domain L1 loss, amplitude contrast loss, and phase contrast loss are used as the main optimization objectives. The loss function is as follows:

[0116]

[0117] In the formula, represents the dehazed image, Y represents the clear image, They represent L1 loss, frequency domain L1 loss, amplitude contrast loss and phase contrast loss respectively. A() and P() represent amplitude and phase information respectively. The overall network loss is expressed as follows:

[0118]

[0119] In the formula, λ 1 ,λ 2 ,λ 3 They are empirically set to 0.1, 0.9, and 0.2, respectively, to balance the dual-domain training.

[0120] like Figure 6 As shown, the present invention provides an image defogging and enhancement system in a dusty and foggy environment of a fully-mechanized mining working face in a coal mine, comprising a camera and a dual-domain coupling image defogging network model module;

[0121] The camera is used to collect dust and fog environment pictures of the fully mechanized mining working face in the coal mine, save them to a local path, and write the foggy image path into a database for image defogging;

[0122] The dual-domain coupled image defogging network model module is used to defog a foggy image to obtain a restored image; an optimized defogging model is obtained by iteratively training the dual-domain coupled image defogging network on the RESIDE-Outdoor dataset.

[0123] The dual-domain coupled image dehazing network is trained as follows: the network is trained for 30 iterations on an RTX 3090, and the training BatchSize is set to 8; in order to achieve better model convergence, the present invention uses the Adam optimizer and sets the parameter β 1 and β 2 Set to 0.9, 0.99; In order to improve the model training efficiency and performance, the present invention adopts the cosine annealing decay strategy to reduce the learning rate from 2×10 -4 Gradually reduce to 1×10 -6 ;

[0124] The trained network model is loaded, the foggy image path is obtained from the database, the foggy image is read, and the defogged image is output after one forward propagation, thus completing the construction of the image defogging enhancement system in the dusty and foggy environment of the coal mine fully mechanized mining working face.

[0125] Example: Comparison of the defogging effect of the method of the present invention and the existing defogging algorithm on the SOTS-Outdoor dataset. The results are as follows Figure 7 (a) to (f). Among them, Figure 7 (a) is the original image to be processed. Figure 7 (b) to (d) are the images processed by the processing methods proposed by Dehame, OKNet, and DeaNet-CR, respectively. It can be seen that there are still gray mottled artifacts and mist in the red rectangular box; Figure 7 (e) is the image processed by the present invention, Figure 7 (f) represents a clear image. Figure 7 As can be seen from the comparison of (f), the method of the present invention generates the most natural image, while retaining more background details, without causing color deviation and edge blur, and the reconstructed image is closer to the real image.

[0126] In order to verify the generalization ability of this method, the present invention collects foggy images in the fully mechanized mining working face environment of a coal mine. Figure 8 (a), it is input into the system of the present invention for processing to obtain a fog-free image, as shown in Figure 8 (b) As shown. It can be seen that the image processed by the present invention has been significantly improved in visual effect. Specifically, the method of the present invention can effectively remove haze caused by factors such as water vapor spraying, and successfully restore the detailed texture information in the image, thereby significantly improving the clarity and overall visual quality of the image.

Claims

1. A method for image defogging and enhancement in a dusty and foggy environment of a fully-mechanized coal mining working face, characterized in that: The steps include: S1. Obtain a foggy image to be processed, and perform an encoding operation on the foggy image through an encoder to obtain encoding features from the foggy image; S2, input the obtained coding features into the bottleneck layer integrated by the DM module to achieve the coupling of the coding features in the spatial domain and the frequency domain to obtain the fusion features; S3. The obtained fusion features are decoded by a decoder and the target defogging image is output.

2. The method for image defogging and enhancement in a dusty and foggy environment of a fully-mechanized coal mining working face according to claim 1 is characterized in that: The specific process of S1 is as follows: S1.

1. Use the RESIDE dataset to obtain foggy images and clear images, randomly crop the foggy images to 256×256, and horizontally flip the foggy images with a probability of 50% to train the defogging network. S1.

2. The dehazing network includes an encoder, a bottleneck layer and a decoder. The encoder performs three downsampling operations to double the feature channel and reduce the width and height to half of the original. Each downsampling operation consists of four ResBlock blocks, and each ResBlock block consists of two 3×3 convolutions and an intermediate GELU activation function repeated four times. The encoding features corresponding to the foggy image features are gradually extracted through the ResBlock blocks of the encoder.

3. The method for image defogging and enhancement in a dusty and foggy environment of a fully-mechanized coal mining working face according to claim 1 or 2, characterized in that: The DM module in S2 includes a convolutional layer, a DCM module, an APGM module and an SRU module; The fusion process of fusion features is: S2.1, through the deformable convolution operator DCM, the spatial context information is effectively aggregated and the texture details of the heavy haze area are reconstructed. For the input features DC-h and DC-w operations are applied to extract context information from the height and width directions respectively. The formula is as follows: In the formula, ω i represents the parameters of the deformable convolution, o c for the width and height dimensions in channel c i The learnable offset on ; S2.

2. Calculate the average value of all pixels in each channel, apply 1×1 point convolution to double the number of channels, and perform softmax operation to obtain the weight ω: ω=softmax(conv(X avg )); Where, X avg Represents X DC-h With X DC-w The average value of all pixels in each channel, conv represents 1×1 convolution, ω represents X DC-h With X DC-w weights array; S2.3, weight ω and X DC-h and X DC-w Multiply and convolve to produce the final output: X DCM =ω[0]*X DC-h ++ω[1]*X DC-w ; In the formula, ω[0] and ω[1] represent X DC-h and X DC-h The weight of X DCM It is the output result of DCM module; S2.4, through the amplitude phase guidance module APGM, the prior knowledge of amplitude and phase is used to guide the extraction of global features: S2.4.

1. For input feature map Apply fast Fourier transform FFT to convert the input features from the spatial domain to the frequency domain, and obtain the amplitude and phase information as follows: phase = atan(B,A); In the formula, is a fast Fourier transformer, A and B represent the real part and the imaginary part respectively, and atan() is the inverse tangent function; S2.4.

2. Perform low-frequency filtering on the amplitude and high-frequency filtering on the phase. Apply a residual block ResBlock consisting of two convolutional layers with a LeakyReLU activation function in the middle to the amplitude and phase components, and then combine the original and modulated amplitude and phase information through an inverse Fourier transform to produce a feature map containing phase and amplitude modulation: with'=conv 1*1 (LeakyReLu(conv 1*1 (with))); phase'=conv 1*1 (LeakyReLu(conv 1*1 (phase))); In the formula, e i represents a complex exponential, It is the inverse operation of the fast Fourier transform; S2.4.

3. Connect the modulated feature maps along the channel dimension, and then perform 1×1 point convolution to reduce the number of channels by half to obtain the final output: X APGM =conv 1*1 (concat(X amb ,X angel )); Where, X APGM Represents the output result of the APGM module; S2.

5. Fusion of spatial and frequency domain features: S2.5.

1. The obtained output features are combined by addition operation to generate reconstruction information combining spatial domain and frequency domain, which is expressed as: F=conv 1×1 (X)++X DCM (X)++X APGM (X); In the formula, conv 1×1 represents the output feature of the convolution operation with a convolution kernel of 1×1, X DCM and X APGM Indicates the output characteristics of the DCM module and APGM module; S2.5.

2. The spatial domain and frequency domain are coupled through the spatial reconstruction module SRU. For the input features First, we use BatchNorm to normalize the weight map, and then use the Sigmoid function to map the weight map to the range (0,1). Then, we perform a soft threshold operation to set the weights above a certain threshold to 1 to obtain information weights; the weights below the threshold are set to 0 to form non-information weights. These weights are multiplied by the input feature F to generate two weighted features, namely, information-rich features and less information-rich features, as shown below: F w =σ(BatchNorm(F)); Where σ is the sigmoid function, F w Represents the weight of input feature F, where F1 and F2 are information-rich features and information-less features respectively; S2.5.

3. Perform reconstruction operation and connect the reconstructed features to generate the final feature map It is expressed as:

4. The method for defogging and enhancing images in a dusty and foggy environment of a fully-mechanized coal mining working face according to claim 3 is characterized in that: The S3 method is as follows: the decoder performs three upsampling operations to reduce the feature channel to half of the original one, and the width and height are doubled; each upsampling operation consists of four ResBlock blocks, each ResBlock block consists of two 3×3 convolutions and an intermediate GELU activation function repeated four times; after the last ResBlock block, the feature is mapped to the RGB space through a 3×3 convolution to output the dehazed image; the specific process includes: S3.

1. The fusion feature is recorded as The encoder is recorded as Decoder, and the dehazed image is obtained after being processed by the decoder It is expressed as: S3.

2. To ensure the consistency between the dehazed image and the clear image at the pixel level and in the frequency domain, L1 loss, frequency domain L1 loss, amplitude contrast loss, and phase contrast loss are used as the main optimization objectives. The loss function is as follows: In the formula, represents the dehazed image, Y represents the clear image, They represent L1 loss, frequency domain L1 loss, amplitude contrast loss, and phase contrast loss respectively. A() and P() represent amplitude and phase information respectively. The overall network loss is expressed as follows: Where λ1, λ2, and λ3 are empirically set to 0.1, 0.9, and 0.2, respectively, to balance the dual-domain training.

5. An image defogging and enhancement system in a dusty and foggy environment of a fully-mechanized coal mining working face, comprising a camera, characterized in that: It also includes a dual-domain coupled image dehazing network model module; The camera is used to collect dust and fog environment pictures of the fully-mechanized mining working face in the coal mine, save them to a local path, and write the foggy image path into a database for image defogging; The dual-domain coupled image defogging network model module is used to defog a foggy image to obtain a restored image; an optimized defogging model is obtained by iteratively training the dual-domain coupled image defogging network on the RESIDE-Outdoor dataset.

Images