127 results about "Vanishing gradient problem" patented technology

The invention discloses an image super-resolution reconstruction method based on deep learning, and belongs to the technical field of image signal processing. Comprising the following steps: S1, realizing multi-dimensional feature representation of a low-resolution input image by using a deep learning network; s2, realizing accurate alignment and lossless mapping of cross-scale features, and effectively solving the problems of feature dimension inconsistency and information distortion; s3, explicitly learning an increment mapping relation between low-resolution input and a reconstructed high-resolution image, and relieving the gradient disappearance problem of a deep learning network structure; s4, enhancing the naturalness and detail fidelity of the reconstructed image through adversarial training; s5, integrating pixel-level reconstruction loss, perception loss and structural similarity loss, and effectively balancing objective indexes and subjective visual quality of the reconstructed image; and S6, realizing joint optimization of deep learning network parameters through a unified gradient back propagation mechanism, and finally outputting a high-quality super-resolution reconstructed image.

The invention provides a rolling bearing fault diagnosis system and method based on a multi-source fusion residual network, and relates to the technical field of bearing fault diagnos.The rolling bearing fault diagnosis system and method adopt a one-dimensional convolution kernel and a residual block feature extraction module, the one-dimensional convolution kernel ensures that data of different sensors are not mixed in the feature extraction process, and the residual block feature extraction module is used for extracting the data of different sensors; the effectiveness of unique features of each type of signals is guaranteed, the gradient disappearance problem in a deep network is relieved through residual connection, a feature fusion module of a grouping cross attention mechanism is further adopted, sensor signals are grouped according to types, each group is weighted independently, and the sensor signals are subjected to clustering processing according to types. And local and global multi-scale features are further extracted by using multi-scale convolution, so that the uniqueness of different sensor signals is ensured, the expression ability of fusion features is improved, and the accuracy and robustness of rolling bearing fault diagnosis are improved.

The invention provides an electrocardiosignal reconstruction method based on hybrid optimization and multi-modal feature fusion, and the method comprises the steps: obtaining a 12-lead electrocardiosignal, processing the 12-lead electrocardiosignal through a linear regression model, a genetic algorithm and a simulated annealing algorithm in sequence to obtain an optimal three-lead electrocardiosignal, and reconstructing the optimal three-lead electrocardiosignal according to the optimal three-lead electrocardiosignal. The optimal three-lead electrocardiosignal is subjected to one-dimensional convolution layer and maximum pooling processing in sequence to obtain time domain features, the time domain features, the frequency domain features and expert features are subjected to multi-modal feature fusion to obtain fusion features, and the fusion features are input into a Transform mechanism to be processed to obtain a reconstructed electrocardiosignal. The method has remarkable effects in long-range time sequence modeling and local waveform detail optimization. The traditional CNN / RNN is limited by the problem of local receptive field or gradient disappearance, the global rhythm and the local form are difficult to balance, the QRS is adopted to perceive a Transform architecture, and a multi-head self-attention and nonlinear feed-forward network is combined, so that the width error of a QRS wave group is smaller.

The invention discloses a road disease detection method based on an unmanned aerial vehicle, electronic equipment and a program product. The method is realized based on a trained road disease detection model, a C3k2-MDDSC module is introduced into a backbone network of the model, the feature multiplexing capability is enhanced through gradient shunting and multi-scale fusion, the gradient disappearance problem is relieved, and the robustness of the model is improved by means of jump connection and packet convolution. An ACFP module is introduced into the tail end of the backbone network, dynamic fusion of local and global features is realized by using multi-scale cavity convolution and a channel-space attention mechanism, and the complex scene modeling capability is improved. And the neck network is integrated with an SGF module, so that the spatial perception of the model to a tiny target can be improved. Besides, the ES-FPN proposed based on the SGF module not only can enhance the utilization of shallow spatial information, but also can optimize the complementarity of cross-level features. During training, regression loss, namely fast high-quality intersection-to-union ratio loss, is proposed, and angle punishment is introduced to improve the alignment precision of the rotating frame and the convergence speed of the model.

The invention provides a multi-channel sequence data-based prediction method and related equipment. The method comprises the following steps: acquiring multi-channel sequence data related to a prediction task specified by a user; the multi-channel sequence data comprises state information of a prediction object associated with the prediction task at a plurality of past time points; performing multi-channel adaptive decomposition operation on the multi-channel sequence data to obtain main frequency components and dynamic characteristics in the multi-channel sequence data; the main frequency components and the dynamic features are input into a prediction model, output data of the prediction model are obtained to serve as a prediction result, and the prediction result comprises state information of the prediction object at multiple time points in the future. According to the method, higher-quality features are provided for the prediction model through multi-channel adaptive Fourier decomposition, the prediction precision can be improved, dependence on a long-time sequence is effectively reduced through combination of the multi-channel adaptive decomposition and the long-short-term memory network, and the gradient disappearance problem in long-time sequence prediction is relieved.

An industrial product cumulative damage image generation method based on an improved DCGAN is characterized by comprising the following steps: step 1, based on a DCGAN model framework, an industrial product cumulative damage image generation network IPD-GAN based on the improved DCGAN is constructed, and the IPD-GAN is provided with a generator and a discriminator; 2, the generator obtains random noise, processes the random noise, generates a damage image and transmits the damage image to a discriminator; step 3, the discriminator obtains a real image, carries out true and false discrimination on the real image and the damaged image, adopts a Wasserstein distance loss function with a gradient penalty term as a model to resist loss according to a discrimination result, guides training of the generator in combination with L1 loss and SSIM loss, and comprehensively optimizes system performance; and step 4, generating a damaged image by using the trained IPD-GAN. The method has the advantages that the problems of unstable training and gradient disappearance existing in the DCGAN are solved.

The invention discloses a method for estimating the energy state of a high-speed lithium battery, which belongs to the field of electric automobiles and comprises the following steps: performing multi-time scale modeling and hierarchical feature extraction on current, voltage and power data of the lithium battery through a deep echo state network, independently adjusting hyper-parameters of each layer by using a parameter differentiation strategy, and calculating the energy state of the lithium battery according to the hyper-parameters; therefore, different dynamic characteristic inputs can be adapted. Furthermore, sudden change output signals in the prediction process are removed through a modal decomposition noise reduction module, and the prediction precision and stability are improved. Experimental verification shows that the method shows high-speed prediction performance, high precision and high usability under various real automobile working conditions, not only has the capabilities of rapid training and efficient prediction, but also can effectively avoid the problem of gradient disappearance in a traditional recurrent neural network, and meanwhile, keeps higher precision and robustness.

An underwater robot application scene-oriented sodium battery residual life prediction method comprises the following steps of: obtaining normalized features after input features pass through a double-path feature normalization module; processing the input features through a multi-local sensing path module, and fusing and splicing the processed input features with the normalized features to obtain spliced features; performing 1 * 1 convolution operation on the input features, adding the input features and the splicing features element by element, and performing processing through a lightweight residual connection module to obtain multi-scale features; enabling the coder-decoder to predict an initial capacity prediction value according to characteristics rich in global time sequence information in the multi-scale characteristic data; performing element-by-element addition on the initial capacity prediction value and the normalized feature, and calculating a final sodium battery residual life prediction value; the coder-decoder can capture the long-range dependency relationship of the battery capacity sequence through a self-attention mechanism, the gradient disappearance problem of the recurrent neural network in the long sequence is solved, and the prediction precision is remarkably improved.

The invention relates to the technical field of network security and abnormal traffic detection, in particular to an SDN abnormal traffic detection model based on an improved bidirectional TCN model, the model takes a traffic quintuple as a basis, extracts a data packet length sequence as a core time sequence feature, constructs an improved bidirectional time convolution network module, and performs joint modeling through a forward TCN and a backward TCN, so that the network security and abnormal traffic detection efficiency is improved. A long-range dependency relationship in a flow sequence is captured, an ELU activation function is introduced into the model to enhance the nonlinear expression ability, the gradient propagation effect is optimized through a residual connection structure, the gradient disappearance problem in deep network training is relieved, and for the problem that correlation modeling among multi-channel features is insufficient, the model is further fused with a multi-head extrusion excitation mechanism, so that the multi-channel feature correlation modeling method is improved. And the importance weight of the feature channel is adaptively adjusted, so that the abnormal traffic detection capability and the model generalization performance are improved.

The invention discloses an underwater bridge crack defect detection method and system based on a U-Net algorithm, and the method comprises the steps: enabling information to be directly transmitted from a shallow layer to a deep layer through quick connection of an added residual network module on the basis of a conventional U-Net algorithm, and effectively solving the problem of gradient disappearance; therefore, the U-Net can construct a deeper network structure, so that richer and higher-level features are extracted, and the ability of understanding complex images is improved. And secondly, the added RFB module can extract scale features of receptive fields of different sizes at the same time, more comprehensive information is provided for the model, and the detection capability of cracks of different sizes is improved.

The invention belongs to the field of deep learning technology and remote sensing image segmentation, and particularly relates to a remote sensing image segmentation method based on Transsubnet edge information enhancement and multi-dimensional feature perception, and the method comprises the steps: S1, preparing a data set; s2, constructing remote sensing picture text description; s3, constructing and training a remote sensing image segmentation model; and S4, storing and testing the model. The invention designs a multi-modal feature extraction method based on parallelism of a sampling branch and a text feature extraction branch under edge feature compensation. The residual error mixing axial attention module is used for forming a transformer structure; and a text-picture multi-dimensional feature fusion enhancement module and a decoder part are embedded. According to the method, the ground feature identification capability can be improved through a text and picture multi-modal feature enhancement strategy, the segmentation boundary and small target object feature information is enhanced, more fine-grained features are reserved, the cross-regional long-distance dependency relationship is better captured, the common gradient disappearance problem in a deep network is relieved, and the method is suitable for large-scale popularization and application. And meanwhile, the small sample data set segmentation effect is improved.

The invention belongs to the technical field of water quality prediction, and provides a multivariable-based water quality prediction method, system, equipment and medium in order to improve the accuracy of water quality prediction. A self-attention mechanism is introduced to solve the problems of gradient explosion and gradient disappearance which are frequently encountered when a traditional recurrent neural network processes a long-time sequence, so that the capturing capability of a prediction model on time sequence characteristics is enhanced; the hyper-parameters are optimized through a particle swarm optimization (PSO) method, so that the prediction precision is improved.

The invention belongs to the field of audio processing, and particularly discloses an audio depression detection method and system based on an improved convolutional neural network autoencoder, and the method comprises the steps: carrying out the potential feature extraction and compression of input original audio data, extracting the data into a low-dimensional and dense potential feature vector, and carrying out the recognition of the potential feature vector; and the information loss in the reconstruction process is reduced. According to the model part, a residual block structure is introduced into an encoder part, input low-layer information is reserved through jump connection, the learning ability of the network is enhanced, and the gradient disappearance problem is avoided. A transpose convolution operation is introduced into a decoder part, the spatial resolution is increased through up-sampling, and a high-quality signal reconstruction task is taken as a constraint, so that the extracted potential features are ensured to contain all key information for accurate classification. Finally, the potential features are directly used for depression classification, and experimental results show that the accuracy and robustness of depression detection can be remarkably improved, and the method has high practical application value.

The invention provides a single-frame infrared small target detection method based on a variant convolutional neural network. The single-frame infrared small target detection method specifically comprises the steps that a picture input module carries out standardized preprocessing and primary feature extraction on an original image; the efficient channel attention module extracts channel descriptors through global pooling, efficiently calculates weights by using a lightweight network, and performs weighted adjustment on input channels; the FPN module realizes efficient fusion of cross-scale features through a space cyclic shift and channel grouping technology; the residual error convolution module adopts a structural design of combining a convolution path and jump connection, so that the problem of gradient disappearance in deep network training is effectively solved; the feed-forward module integrates an efficient channel attention mechanism, and enhances the response capability of a key feature channel through global information aggregation, channel weight generation and re-calibration operation; the model output module converts high-level features into high-precision mask images and calculates detection performance indexes in real time, all the modules work cooperatively to form a complete feature extraction-fusion-prediction processing chain, and the performance of the infrared small target detector is remarkably improved while the real-time reasoning efficiency is guaranteed.

The invention provides a pavement well lid defect automatic detection method and system based on an improved YOLOv8 model, emphatically optimizes a model architecture, and aims to improve the detection precision and calculation efficiency of pavement well lid defects. According to the improved YOLOv8 model, a Mish activation function is adopted to replace an original SiLU activation function, and a CGA attention mechanism is introduced. The use of the Mish activation function alleviates the gradient disappearance problem to a certain extent, so that the information can better flow in the network, and the stability and effectiveness of network training are promoted. The CGA attention mechanism can enable the network to pay more attention to important parts during feature processing, and the capacity of capturing well lid defect features is enhanced. Through the improvements, the expression ability of the model is improved. The improved YOLOv8 model not only can realize efficient calculation, but also shows higher precision in automatic detection of the well lid defects on the road surface, and provides a more advantageous solution for well lid defect detection.

The invention relates to a photoetching hot spot detection method based on an Inception block and a residual network. The invention provides a data enhancement and preprocessing method, a photoetching pattern is processed firstly, a training data set is effectively expanded through the step, a data enhancement processing method is adopted, a hot spot pattern is subjected to 180-degree rotation, mirroring and 180-degree rotation and then mirroring, and the recognition capability of a model on the hot spot pattern is enhanced. On the basis, the invention provides a deep learning network adopting an Inception-Resnet neural network structure to solve the photoetching hot spot detection problem, and the neural network combines the multi-scale feature extraction capability of an Inception module and the residual connection of Resnet, so that the network can effectively solve the gradient disappearance problem while keeping the depth, and the detection accuracy is improved. And the training stability and the model learning efficiency are improved.

The invention discloses a bearing fault diagnosis method and system based on a multi-source fusion deep adversarial network, and belongs to the technical field of fault diagnosis. According to the method, an improved normalized Hilbert-Huang transform time-frequency spectrum feature extraction model is introduced to capture local features of a fault data time-frequency domain, and pooling operation is adopted to compress two-dimensional time-frequency image features in fault data time-frequency information so as to highlight important features. And a splicing operation is adopted to obtain a multi-source signal fusion feature result so as to capture multi-scale fault features. An improved Wasserstein generative adversarial network model is adopted to obtain data distribution results of real samples of different fault categories, so that a comprehensive data distribution model can be constructed, and the generalization ability of a fault diagnosis system can be improved; and an enhanced deep residual shrinkage network model is adopted to solve the gradient disappearance problem in the deep network, and the shrinkage network is adopted to improve the sparsity of the weight of the deep network to improve the generalization of the model.

The invention discloses a data set distillation method and system based on semantic relation decoupling, and belongs to the technical field of data distillation. According to the method, the feature distribution of the distillation data is aligned with the category-level statistics of the original data stored in the memory bank to realize efficient and fine-grained category-level optimization of the distillation data, so that the representativeness and semantic consistency of the distillation data are improved; in addition, for the problem that distribution in the last layer of feature distribution of the pre-training model is too concentrated, the problem of downstream task gradient disappearance caused by insufficient sample diversity in the feature space is effectively relieved by maximizing diagonal elements of a covariance matrix and minimizing non-diagonal elements at the same time. Therefore, the generalization ability and the learning effect of the model on downstream tasks are improved.

The invention belongs to the technical field of deep learning, and particularly relates to an ozone concentration prediction method based on a CNN-AM-BiLSTM residual network. Comprising the following steps: acquiring historical conventional pollutant and meteorological factor data of a monitoring station and carrying out data cleaning; carrying out feature extraction on the input multi-dimensional data by the convolutional neural network; a multi-head attention mechanism is constructed to help the model to better pay attention to important features, so that the feature representation capability is enhanced; the BiLSTM performs time modeling on the extracted features, so that the long-term dependency relationship in the time sequence data is better captured; a residual network is constructed to relieve the gradient disappearance problem, so that the model better learns nonlinear mapping of input data; the characteristics output by the BiLSTM are mapped to the target output space through the full connection layer, the invention provides a practical and effective method for ozone concentration prediction, the ozone concentration can be accurately predicted, and effective technical support is provided for atmospheric environment treatment.

The invention relates to the technical field of building image processing, in particular to a high-spatial-resolution remote sensing image building extraction method. Comprising the following steps: firstly, preprocessing an original remote sensing image; and then, a BuildingNet model is introduced to extract the building. All the preprocessed samples are used as the input of the model, and the trained model outputs two types of classification diagrams. A multi-layer dense connection convolution block adopted by the BuildingNet model is excellent in performance in image classification and has the advantages that the gradient disappearance problem in a deep neural network is relieved, the feature extraction capacity is high, feature propagation in training and evaluation is promoted, feature reuse in classification and segmentation tasks is encouraged and the like, and meanwhile, the DenseNet can reduce the number of parameters, so that the method is easy to train.

The invention provides a network traffic matrix prediction method based on multi-scale convolution and attention collaborative enhancement LSTM (Long Short Term Memory), and solves the problems that an existing traffic matrix prediction model is insufficient in multi-scale feature capture, weak in space-time correlation modeling and poor in generalization. A dynamic multi-scale convolution module (adopting a 3 * 3 / 5 * 5 / 7 * 7 two-dimensional convolution kernel) is used to extract local fine-grained association and global coarse-grained trend of a two-dimensional traffic matrix, deep fusion of LSTM time sequence features and two-dimensional multi-scale spatial features is realized in combination with a cross attention mechanism, and time-space cross-dimension association is enhanced. And meanwhile, the problem of gradient disappearance of the deep network is relieved by relying on residual connection, and topology independence is realized based on a flow-by-flow method so as to adapt to traffic matrixes of different dimensions. The method specifically comprises the following steps: performing stream-by-stream division and normalization preprocessing on an original two-dimensional traffic matrix, and after processing by modules such as dynamic multi-scale convolution, cross attention enhancement LSTM, residual block and gating screening and the like, outputting a single OD stream predicted value and reconstructing a complete traffic matrix; and integrating objective functions of the encoder and the decoder, taking the MSE as a loss function, and adopting an Adam optimizer to iteratively update model parameters until the model converges. Experiments show that the MAE and the MSE of the method on real data sets such as Ailene, GEANT and the like are better than those of mainstream models such as SVR, LSTM, MTGNN and the like, the reconstruction precision of a high-dimensional traffic matrix is particularly remarkable, and a reliable basis can be provided for network congestion control and resource optimization.

The invention discloses a nuclear power material purchase demand prediction method and system based on a time convolutional network, and relates to the technical field of time series data analysis, and the method comprises the steps: collecting a data set, carrying out the preprocessing of the data set, classifying the data set, setting a representation mode of the input data set, and constructing a time convolutional neural network model. A data set is divided, model training is performed by using the divided data set, time sequence data is input into a trained model, and a material purchasing demand prediction value is output through multi-layer convolution, residual connection and full connection layer processing. According to the method, the time sequence data features are deeply learned through expansion convolution and residual connection, the ability of capturing the complex time dependency relationship by the model is improved, gradient flow in a deep network is promoted through introduction of the residual connection, the gradient disappearance problem is avoided, and the training efficiency and the learning ability of the complex features of the model are improved.

The invention relates to an interactive hybrid quantum-classical machine learning model training method and device and a medium, and the method comprises the steps: loading training data, and initializing a classical neural network and a quantum machine learning model; inputting the classic data into a classic neural network to obtain a classification result as prediction output; inputting the quantum data into a quantum machine learning model, taking the output of the classical neural network as a parameter of quantum machine learning, and obtaining an expected value of a quantum bit under considerable measurement as a prediction output; calculating a weighted joint loss function based on a classic neural network, prediction output of quantum machine learning and a true value of a data set; updating classic neural network parameters based on a weighted joint loss function; when the weighted joint loss function converges or reaches the maximum number of iterations, training is ended. Compared with the prior art, the method has the advantages that the expense of quantum machine learning training is reduced, and the potential gradient disappearance problem of a quantum machine learning algorithm is solved.

The invention discloses a two-stage brain glioma segmentation system based on an improved U-Net model, and belongs to the field of medical image segmentation. According to the method, the problems of low precision, low efficiency and poor generalization ability of the existing segmentation technology are solved. In a tumor tissue detection stage, the lightweight neural network is utilized to realize preliminary identification of a brain glioma region, and the calculation range of subsequent fine segmentation is effectively reduced, so that the model can perform reasoning at relatively low resource and time cost, and the segmentation efficiency is improved. By adopting the multi-scale feature residual module, the problem of gradient disappearance caused by network deepening can be relieved, information can be propagated in a deeper level in the network, and extra parameter and calculation complexity overhead cannot be brought. Meanwhile, jump connection is changed into a multi-scale jump connection structure, the problem that a decoder network can only receive same-scale feature information is solved, and the tumor segmentation precision is remarkably improved. The method can be applied to medical image segmentation.

The invention discloses a brain load assessment method and system based on BEAM. The method comprises the steps that EEG data sets corresponding to high, medium and low symbol loads of a subject are collected, and labels are edited; carrying out data preprocessing on the collected data to reduce the calculation amount and reserve effective frequency band information; calculating the power spectrum density by adopting a Welch method; constructing a color BEAM image by using the obtained power spectrum density and the electrode coordinates; inputting the BEAM sequence into a convolutional neural network to extract deep spatial features; the image sequence is transmitted into a BiLSTM module to capture time sequence features; constructing a GUI (Graphical User Interface) to visually display the spatial distribution condition of the electroencephalogram signals; compared with the traditional electroencephalogram frequency band feature vector research, the method has the advantages that the position information between the electrodes can be better reserved; the gradient disappearance problem in the deep network is effectively solved; the prediction accuracy is well improved; the active condition of each functional area of the brain is visually displayed, and the convenience of man-machine interaction is improved.

The present invention proposes a method for reconstructing an ECG signal based on hybrid optimization and multimodal feature fusion, the method comprising: obtaining a 12-lead ECG signal, processing the 12-lead ECG signal in sequence through a linear regression model, a genetic algorithm, and a simulated annealing algorithm to obtain an optimal three-lead ECG signal, processing the optimal three-lead ECG signal in sequence through a one-dimensional convolutional layer and a maximum pooling layer to obtain time domain features, fusing the time domain features, frequency domain features, and expert features into a multimodal feature to obtain fused features, and inputting the fused features into a Transformer mechanism for processing to obtain a reconstructed ECG signal. The present invention achieves significant results in long-range time series modeling and local waveform detail optimization. Traditional CNN / RNN is limited by the local receptive field or gradient vanishing problem, making it difficult to balance the global rhythm and local morphology. The present invention adopts a QRS-aware Transformer architecture, combined with multi-head self-attention and a nonlinear feedforward network, to reduce the error in the width of the QRS complex.

The invention discloses a remote sensing image road extraction method, system and device based on an improved Transform network and a storage medium, and relates to the technical field of road extraction. According to the technical key points, the method comprises the following steps: carrying out road extraction on a road remote sensing image by utilizing a road extraction model based on an improved Transform network, wherein the improved Transform network comprises a cross-scale coding layer, an improved encoder and a decoder; the cross-scale coding layer solves the influence of Transform multi-scale input on the road extraction effect in the coding stage; a multi-head attention mechanism is replaced by a regional multi-head attention mechanism with a residual term in an encoder and a decoder, so that not only can the features of small objects be accurately concerned and learned, but also more original features can be reserved, and the gradient disappearance problem can be relieved; and the feedforward neural network is improved by using a gating mechanism, so that the expression ability of the model is enhanced. According to the method, road detail information can be extracted more accurately, and a new solution is provided for automatic extraction of road features in high-resolution remote sensing images.

The invention provides a fan blade crack detection method, which comprises the following steps of: for a fan blade image acquired by an unmanned aerial vehicle, performing deblurring processing on the image through a pre-processing algorithm designed by the invention, and overcoming the problem of motion blurring when the unmanned aerial vehicle acquires the fan blade image; and carrying out crack detection on the acquired image by adopting a crack detection neural network model improved based on YOLOv8. According to the crack detection model, self texture interference of the blade can be avoided, the crack detection precision is improved, crack characteristics of different scales are captured through a Stem module of a multi-branch structure, a CAM module is added into a C2f-1 module, crack attention is integrated, and crack related characteristics are enhanced; a residual connection mode of a residual block is designed in a C2f-2 module to avoid the problem of deep network gradient disappearance, gradient flow and effective transmission of crack characteristics are guaranteed, response of crack-related channels is enhanced through channel re-calibration, and irrelevant channels such as blade textures are inhibited.

The invention discloses a mine dust concentration image recognition method based on prior features and a multi-kernel residual attention network, and the method comprises the steps: collecting mine dust images under different concentration conditions, extracting multi-dimensional image features, and constructing a prior feature library; linear and nonlinear feature subsets significantly related to the mine dust concentration are screened out; respectively mapping the linear and nonlinear feature subsets to a high-dimensional space by adopting a multi-kernel learning framework and carrying out weighted fusion to obtain fused high-order features; a multi-kernel residual attention network is constructed and trained, the gradient disappearance problem is relieved through a residual block structure, the weight of each feature channel is learned through a channel attention mechanism, and dynamic enhancement of key mine dust image features is carried out; and inputting the fused high-order features into a multi-kernel residual attention network for regression calculation, and outputting a mine dust concentration predicted value. According to the method, high-precision, non-contact and high-interpretability image recognition of mine dust concentration can be realized, and the method is suitable for real-time mine safety monitoring in a complex environment.

This invention discloses a gene regulation inference method guided by topological data analysis and gene network embedding. It combines TDA and GNN to enhance the inference capability of gene regulation networks. By capturing the topological structure of the gene regulation network graph through TDA features, the model's ability to model gene expression is enhanced. The TDA features and GAT embedding representations are effectively integrated through gating fusion. This fusion mechanism enables the model to adaptively adjust node embeddings based on global topological characteristics, which not only improves the accuracy of gene interaction representation but may also enhance the accuracy of regulatory relationship prediction. The traditional GAT architecture is extended through a four-layer graph attention mechanism. Each layer uses residual connections to alleviate the gradient vanishing problem and improve training stability. In addition, independent multilayer perceptron branches are designed for transcription factors and target gene embeddings. This deep architecture can achieve more expressive feature transformations and capture subtle patterns in gene regulation networks.