81 results about "Sigmoid activation function" patented technology

The invention provides a dual-binary Turbo decoding method based on deep learning, and belongs to the technical field of wireless communication. Comprising the following steps: generating a binary random bit stream as original information, and generating a Turbo code sequence through two recursive system convolutional code component encoders; pre-processing the Turbo code sequence, and inputting the pre-processed Turbo code sequence into a bidirectional PNN model for processing to obtain an output feature vector; based on the output feature vector, standardization is carried out through a batch normalization layer, and a decoding result probability value is output by using a full connection layer and a Sigmoid activation function; and on the basis of the decoding result probability value, a Max-Log-MAP algorithm is cascaded with a component decoder of the bidirectional RNN to obtain a decoding result.

The invention discloses a method for predicting an RNA (Ribonucleic Acid) methylation site, which is based on a multi-modal feature fusion and semantic vector embedding technology and is used for remarkably improving the recognition precision of an m7G modification site. The method comprises the following steps: firstly, constructing an RNA sequence data set containing positive and negative samples, and dividing the RNA sequence data set into a training set and an independent test set according to a predetermined proportion; then, multi-modal features are extracted through One-hot coding (One-hot), nucleotide chemical property coding (NCP), electron-ion interaction potential coding (EIIP) and local nucleotide composition coding (ENAC), and context semantic information of a nucleotide sequence is obtained in combination with a DNA2Vec model; according to the model, a multi-modal feature fusion path (MRF) and a DNA2Vec embedding path are adopted, after feature dimension compression is carried out through a full connection layer, a Transform encoder is used for capturing a long-range dependency relationship, and a prediction probability is calculated through a Sigmoid activation function. In the optimization process, batch normalization, Dropout and Adam optimizers are adopted, and the binary cross entropy loss function is minimized. Finally, the performance of the model is evaluated through five-fold cross validation, and the generalization ability of the model is verified on an independent test set. According to the method, through multi-source feature fusion and hierarchical modeling, the analysis capability of sequence information is remarkably improved, and an accurate calculation tool is provided for RNA modification prediction.

The invention discloses a GRU-based vehicle-mounted CAN intrusion detection method and system, a storage medium and a computer system. According to the method, an automatic encoder (AE) is introduced to deepen the understanding of a model on input sequence characteristics, a sliding window is used for selecting batch CAN data to be preprocessed to obtain 13-dimensional time sequence data, and a scalar value within the range of [0, 1] is obtained through processing of the encoder, a GRU, a decoder, a full connection layer and a sigmoid activation function and used for classification of abnormal data. The Conv1D is used as a hidden layer, and compared with two-dimensional convolution, the one-dimensional convolution parameter quantity is smaller, and the calculation is simpler and more convenient. An attack message and a normal message can be completely distinguished, the precision and the accuracy rate reach 100%, and the precision in Fuzz detection is 0.9983; compared with the prior art, the method has high accuracy and reliability in the aspect of intrusion behavior detection, can effectively identify most intrusion events, and can keep a relatively low overall error rate, so that good balance between safety and availability is realized.

The invention relates to an underground coal mine robot path planning method based on deep reinforcement learning, and the method comprises the steps: obtaining to-be-predicted state information, designing a bimodal reward function, and driving a robot to move towards a target position to avoid an obstacle; inputting the to-be-predicted state information into the path planning model to obtain a path planning result; the path planning model is obtained by training the improved DDPG double-layer network model by using a training set; the step of improving the DDPG double-layer network model comprises the steps that an LSTM layer is added in front of an Actor network module of the DDPG double-layer network model and used for making full use of environment information, a full connection layer in a hidden layer of the Actor network module adopts a ReLU activation function, a Sigmoid activation function is introduced into the last layer of the Actor network module, it is ensured that output values are all non-negative values, and the LSTM layer is used for making full use of environment information. And replacing the last full connection layer of the Critic network module in the DDPG double-layer network model with the improved crevasse network.

The invention discloses a training method, a segmentation method and a device of a multi-mode weak supervision medical image segmentation model. The method comprises the following steps: unifying the size of a medical image; in the first stage of the network model, CAM graphs under different scales are extracted through a residual coding network, and single-mode CAM graphs under different scales are merged through a Sigmoid activation function; fusing CAMs of different modes; encoding a foreground image, a background image, a text label corresponding to the foreground image and a text label corresponding to the background image by using CLIP, calculating the similarity between an image encoding vector and a corresponding text encoding vector, and maximizing the similarity as a loss item; meanwhile, taking an opposite number of an absolute value of a difference value between the CAM of the foreground image and the CAM of the background image as one of loss items to train a model; performing foreground and background segmentation on the final CAM by taking 0.5 as a threshold value, and calculating a final Dice score with a real label so as to evaluate the performance of the model; according to the method, the multi-modal CAM fusion problem can be solved by utilizing the strong representation capability of the large model.

The invention discloses a remote sensing image segmentation method based on Gabor transformation and orthogonal attention, and the method comprises the steps: carrying out the feature extraction of a preprocessed target remote sensing image through the combination of Gabor transformation and a deep learning model, and obtaining an initial feature; processing the initial features based on an orthogonal channel attention mechanism to obtain channel attention; performing element-by-element multiplication on the channel attention and the initial feature to obtain a weighted feature map; performing convolution on the initial features through a dynamic separable convolution layer to obtain global information; extracting global information by adopting a global average pooling layer, and generating a weight map through a second Sigmoid activation function; performing full-dimensional dynamic convolution on the initial features, and multiplying the initial features with the weight map to generate optimized features; fusing the weighted feature map and the optimized feature to obtain an enhanced feature map; and converting the enhanced feature map into an image segmentation result. The method can improve the precision and robustness of remote sensing image segmentation in a complex scene.

The invention provides a streaming speech recognition method. The streaming speech recognition method comprises the following steps: S1, collecting an audio data set; s2, performing feature extraction on data in the preprocessed audio data set to obtain audio features; s3, inputting the audio features into a trained encoder for encoding, and outputting encoded and normalized feature representation; s4, the encoded and normalized feature representation is fed into a trained CIF module, the CIF module uses a one-dimensional convolution layer to generate a weight, and meanwhile, Dropout and sigmoid are used to activate functions, and integrated acoustic features are output; and S5, feeding the integrated acoustic features to a trained decoder for decoding to obtain a prediction probability. According to the improved Conformer streaming speech recognition method based on the neuron integration emission mechanism, the recognition accuracy is improved.

The invention relates to the technical field of geophysics, in particular to an electromagnetic method three-dimensional inversion method and system based on a convolutional neural network, and the method comprises the following steps: firstly, achieving the end-to-end nonlinear mapping through a three-dimensional CNN processing module, and obtaining a corresponding three-dimensional resistivity model; the three-dimensional CNN processing module comprises an encoder used for extracting multi-scale features through a 3D convolution layer and down-sampling, a decoder used for recovering spatial resolution through a 3D deconvolution layer and jump connection, and a processor used for processing the multi-scale features through the decoder. And the output layer is used for restraining the resistivity range through a Sigmoid activation function. And then multi-frequency electromagnetic field data is used as joint input of a three-dimensional resistivity model, and TV regularization and physical property range constraint are introduced into a loss function to realize three-dimensional inversion. According to the method, efficient, high-precision and low-cost electromagnetic method three-dimensional inversion is realized through a three-dimensional CNN architecture, multi-frequency data fusion and physical constraint design, and the method is remarkably superior to a traditional iteration method.

This application provides a joint emotion recognition method and system based on action unit-driven attention, relating to the fields of computer vision and affective computing. The method includes: inputting a feature map into an action unit (AU)-driven attention branch, processing it through a sigmoid activation function to obtain an activation vector, and inputting this vector into a learnable mapping layer; projecting the AU information back into the spatial geometric space to generate a single-channel spatial attention map, which is then multiplied element-wise with the feature map to obtain a weighted feature map; determining continuous values ​​for discrete emotion category probabilities, valence, and arousal; constructing a total loss function; and training the model using the acquired AffectNet and Aff-wild2 datasets to determine the target model for emotion recognition. This application achieves end-to-end joint optimization of AU detection and emotion recognition, significantly enhancing the interpretability and generalization ability of the model while improving emotion recognition accuracy, enabling it to better adapt to the emotion recognition needs in complex scenarios.

The application relates to the technical field of deep learning and recommendation system, and particularly discloses a hybrid expert model sparse inference method and system for generative recommendation, which comprises the following steps: obtaining original input data, and obtaining an input vector through an embedding layer; obtaining hybrid expert weights through normalization and maximum value selection operation on the input vector; calculating attention-enhanced features according to the hybrid expert weights based on a hierarchical attention mechanism; inputting the input vector and the attention-enhanced features into a hybrid expert model to calculate a recommendation result; wherein, according to the attention-enhanced features, expert weights are calculated based on a parallelized gating mechanism activated by a Sigmoid activation function; the activated expert layer is selected according to the expert weights, and the input vector is used to calculate the recommendation result. The application can achieve a recommendation accuracy comparable to or even higher than that of an advanced dense model under low time overhead and low calculation complexity.

The present invention relates to an underwater image enhancement method based on adaptive multi-scale fusion and attention mechanism. The underwater image enhancement model used includes an encoder, a decoder, and a bottleneck layer located between the encoder and the decoder. The encoder includes a depth convolution layer, an adaptive multi-scale fusion module, and downsampling. The adaptive multi-scale fusion module is used to extract features of different scales and fuse the feature maps of different scales. The decoder includes an improved channel-spatial attention module, upsampling, and a depth convolution layer. The input feature map of the adaptive multi-scale fusion module passes through two dilated convolution layers to obtain two feature maps of different scales. After the two feature maps are concatenated in channels, they pass through a convolution layer, then through a fully connected layer and a Relu activation function, and then through another fully connected layer and a Sigmoid activation function to obtain two weights of different scales. The feature maps extracted by the two dilated convolution layers are multiplied by the weights of the corresponding scales respectively and then concatenated in channels to obtain the output feature map of the module. This method effectively solves the problems existing in underwater images, such as color distortion, low contrast, and blurred details.

The invention discloses a hierarchical adaptive multi-modal fusion method and system for a multi-modal financial large model, and the method comprises the steps: collecting financial text data, financial image data and time series data, and carrying out the preprocessing; performing modal feature extraction including text features, image features and time sequence features, and generating a text feature vector, an image feature vector and a time sequence feature vector; designing a full connection layer for each modal feature, then connecting a Sigmoid activation function, generating a gating weight vector, and obtaining a weighted feature vector; carrying out information interaction between modes by adopting a multi-head attention mechanism to form a feature vector after interaction enhancement; and constructing a financial domain knowledge graph, generating a domain knowledge vector, and carrying out vector splicing on the feature vector after interaction enhancement and the domain knowledge vector to generate a fusion feature vector. The method aims at improving the understanding and analysis ability of a large language model in application scenes such as financial analysis, risk assessment and market prediction.

The invention discloses a weak light image enhancement method based on Transform. The method comprises the following steps: extracting local detail information of an input weak light image through a convolutional layer, and converting the local detail information into a local feature map with a specified size; inputting the local feature map into three continuous feature extraction modules constructed based on 3D multi-head attention to extract deeper global features; fusing the three global features through a multi-layer fusion module to generate a final enhanced feature map; the multi-layer fusion module fuses features from different layers by adopting an adaptive weighting method; and the enhanced feature map passes through a convolutional layer and a Sigmoid activation function to obtain a final enhanced image. According to the invention, the quality of an image shot under a low-light condition is improved.

The invention discloses a vehicle re-identification method and device, equipment and a medium, relates to the technical field of computer vision, and aims to solve the problem that attribute information of a vehicle cannot be utilized due to lack of data annotation by extracting a semantic feature vector related to the color of the vehicle and a semantic feature vector related to the type of the vehicle. The vehicle color features and the vehicle type features are aligned to the spatial dimension of the vehicle global features, the two aligned features are spliced into attribute features, and the attribute features pass through a convolutional layer and a sigmoid activation function to generate a spatial adaptive color space attention map and a spatial adaptive type space attention map respectively; in the process, the weight of the vehicle color feature and the weight of the vehicle type feature are adaptively and dynamically allocated to improve the feature representation capability, and finally, the global feature capable of representing the appearance feature and the visual feature is spliced with the two weighted features after weight adjustment to perform vehicle re-identification, so that the accuracy of vehicle re-identification is improved.

The present invention discloses a forest fire prediction method and system based on an XGBoost-AttentionBiLSTM model. The method includes: S1, constructing a forest fire prediction sample data set; S2, constructing an XGBoost-AttentionBiLSTM model and using the forest fire prediction sample data set for model training and learning; S3, an other factor data processing module extracts terrain factor data and vegetation factor data using terrain data and vegetation data, and the XGBoost layer outputs the wildfire occurrence probability through a sigmoid activation function; S4, collecting meteorological data, terrain data, and vegetation data containing spatio-temporal information at the prediction time and obtaining historical meteorological data before the prediction time, inputting them into the XGBoost-AttentionBiLSTM model, and obtaining the wildfire occurrence probability. The present invention can obtain the wildfire occurrence probability at each position point at the prediction time, and can also output the early warning wildfire points and the corresponding wildfire occurrence probabilities, which is convenient for timely preventive treatment of the early warning wildfire points.

This invention discloses a medical image segmentation method based on enhanced attention. The method acquires a medical image, preprocesses it, and inputs it into an encoder to output a multi-scale feature map of the medical image. The encoder includes N cascaded multi-dimensional attention modules. Each multi-dimensional attention module extracts multi-scale features through dual-channel convolution and residual connections, and performs feature recalibration using an attention enhancement module. The output multi-scale feature map is then fused by a multi-scale dilated fusion attention module to obtain the feature map output by the encoder. The feature map output by the encoder is input to the decoder, where it is combined with high-resolution feature maps of the corresponding levels passed from the encoder via skip connections. Through progressive upsampling and feature fusion, a reconstructed feature map is obtained. The reconstructed feature map output by the decoder is used to generate a medical image segmentation mask through 1×1 convolution and a sigmoid activation function to complete the segmentation of diseased tissues. This invention significantly improves the boundary segmentation accuracy for knee joints, breast tumors, and skin lesions.

The invention relates to the technical field of indoor positioning, and discloses an indoor multi-target positioning method and positioning system based on an antenna array, and the method comprises the steps: collecting the radio frequency signals of a region from a plurality of space angles, and obtaining the RSSI and phase of each tag; preprocessing the RSSI and the phase, inputting the RSSI and the phase into a pre-trained double-branch residual network architecture for feature extraction and feature fusion, and generating a fusion feature map; inputting the fused feature map into a network architecture through a data processing unit to carry out deep feature extraction, and determining output features; and mapping the output feature into an N-dimensional grid position vector, and outputting the probability that each grid in the N grids has a target by adopting a Sigmoid activation function. According to the embodiment of the invention, the RSSI data and the phase data of the reference tag array are acquired from a plurality of space angles through the plurality of virtual antennas in the RFID equipment, a large amount of multi-dimensional radio frequency information is acquired, the space sensing capability of a positioning area is enhanced, and the accuracy of indoor multi-target positioning is improved.

The present invention discloses an image fusion method and system based on a dual-discriminator generative adversarial network, belonging to the field of computer vision. The method includes constructing and training a dual-discriminator generative adversarial image fusion model; wherein the generator includes a dense feature extraction and fusion module, a feature enhancement module, and a decoding and reconstruction module; the dense feature extraction and fusion module extracts and fuses the features of the two frames of images to be fused to obtain a fused feature map; the feature enhancement module performs a global average pooling operation, two fully connected layers, and a Sigmoid activation function on the fused feature map to obtain a feature enhancement coefficient, and then multiplies the feature enhancement coefficient with the fused feature map to obtain an enhanced feature map; and the decoding and reconstruction module decodes and reconstructs the enhanced feature map to obtain a fused image. The present invention effectively retains the information in different source images, while improving the generator's ability to fit and model the fusion of different source images, thereby improving the quality of the fused image.

The invention discloses an airway segmentation network based on multi-scale directional attention and local graph convolution fusion, and relates to the field of medical image processing, a U-Net type three-dimensional encoder-decoder structure is adopted, space detail information is reserved and gradient propagation is promoted through jump connection, an encoder is used for extracting multi-level semantic features, and the decoder is used for extracting multi-level semantic features; a decoder gradually recovers spatial resolution through jump connection, a segmentation result is obtained through a convolution layer and a Sigmoid activation function, an airway segmentation network integrates a multi-scale direction attention mechanism and a local graph convolution attention mechanism, enhancement is carried out in combination with relative position coding, the identification capability of small airway branches is improved, and boundary false detection is reduced. According to the method, the recognition rate of the tail end fine bronchus is remarkably improved, breakage and leak detection are reduced, the connectivity and topological integrity of the airway tree are enhanced, the model is more sensitive to feature response of small airways, fuzzy boundaries and low-contrast areas, the continuity of the output airway model is higher, and the method can be directly used for bronchoscope robot navigation and path planning.

The invention discloses a data classification method and device based on linear feature enhancement, equipment and a storage medium. The method comprises the following steps: respectively carrying out linear propagation on node features and corresponding neighbor node features to obtain a first linear feature and a second linear feature; performing nonlinear mapping on the first linear feature and the second linear feature through KAN nonlinear mapping to obtain a first mapping node feature and a second mapping node feature; determining the embedding of the first node in each layer of the graph convolutional network by combining the first mapping node feature and the second mapping node feature through the graph convolutional network; superposing the embedding of each layer to obtain a target node feature representation of the first node; based on the Sigmoid activation function, the # imgabs0 # activation function and the target node feature representation, determining an enhancement feature of the first node; and converting the enhanced features through a softmax function to obtain a classification result. The method effectively guarantees the calculation efficiency of graph data classification, and improves the classification capability of complex graph data.

This invention claims protection for a CNN-LSTM algorithm neural network accelerator based on FPGA. The CNN hardware implementation includes a data input line buffer module, a convolution calculation module, a ReLU activation function module, an intermediate result buffer module, and a pooling calculation module. The LSTM hardware implementation includes an LSTM control module, a gate function calculation module, and a sigmoid activation function linear approximation module. The FC hardware implementation includes an FC control module, a fully connected layer calculation module, a ReLU activation function module, and a data output buffer. The purpose of this invention is to design a high-performance, low-power, and highly flexible CNN-LSTM neural network accelerator tailored to specific application scenarios. The innovation lies in the fact that, compared to traditional neural network accelerators, this invention uses a parallel pipelined design method to implement a CNN-LSTM algorithm neural network accelerator, which significantly improves the low power consumption and data throughput of the neural network accelerator. Furthermore, the parallel processing capabilities of the FPGA enable the algorithm to run at a faster speed.

This invention discloses a method and system for analyzing and outputting TCM acupuncture cases based on machine learning, belonging to the field of medical image processing technology. The method acquires facial image sequences from two pathological cycles of the subject, extracts spatiotemporal feature maps, and obtains local receptive field feature matrix pairs through sliding windowing. It calculates the cross-cycle joint entropy gradient and its matrix, and generates a two-dimensional deformation vector using optical flow. Using the coordinates of the first cycle as a reference, the deformation vector is set to zero when the joint entropy gradient does not exceed the rigidity threshold; otherwise, the joint entropy gradient matrix is ​​used as a weight matrix through a sigmoid activation function, and a Hadamard product is performed with the deformation vector to obtain a denoised deformation vector. The denoised vectors are superimposed to obtain corrected pixel coordinates, which are mapped to the target localization coordinate matrix, and the acupuncture case analysis results are output. This scheme achieves precise spatial decoupling of local tissue evolution, eliminates the contamination of rigid regions by global deformation, and significantly improves the robustness and accuracy of cross-cycle acupuncture target localization.

The invention discloses a micro-expression recognition method based on a position-motion double-flow fusion network. The micro-expression recognition method comprises two parts: designing a position extraction branch PEM based on Vision TransformerVIT; designing a motion feature extraction branch based on dynamic cross attention; a depth separable convolutional layer is embedded in front of the multi-head self-attention MSA module. A spatial adaptive feed-forward network is designed to replace a multi-layer perceptron in a standard VIT. A grouping processing strategy is adopted, and input features are divided into a plurality of sub-feature groups according to channel dimensions. A parallel sub-network is constructed to focus on feature extraction of different dimensions. And finally, generating a gating signal through a Sigmoid activation function by the interacted features, and performing element-by-element multiplication on the gating signal and the original input features. Through the dynamic weighting mechanism, which key features are reserved and which irrelevant noise is inhibited can be adaptively determined, so that accurate extraction of weak motion features is realized.

The invention relates to the technical field of computer vision, in particular to a banana maturity identification method based on deep learning. The method comprises the following steps of S1, collecting a banana image data set, and performing preprocessing; s2, improving a Si gmoid activation function in the polarization self-attention mechanism module, and introducing the improved polarization self-attention mechanism module into a backbone network of a YOLOv8n model; s3, inputting the pre-processed picture into the improved YOLOv8n network, and inputting the pre-processed picture into the improved YOLOv8n network; and S4, outputting the recognition result graph, the recognition precision and the parameter quantity of the improved model. According to the banana maturity identification method based on deep learning provided by the invention, the maturity of different banana fruits can be rapidly and accurately identified, so that the grading speed of bananas in the transportation and sales process is greatly improved.

The invention provides a load data missing value dynamic filling method and device, equipment and a storage medium. The method comprises the steps that firstly, a real-time power grid parameter input vector of a clean heating transformer area is acquired, a space estimation result is generated through a KNN algorithm, and a time sequence prediction result is generated through an LSTM model; and the meta-model generates output through a full-connection ReLU hidden layer and transmits the output to an output layer, and a dynamic fusion weight is generated through Sigmoid activation function processing. And fusing the results to generate a final filling value, and inputting the final filling value into a load prediction model. According to the method, the KNN and LSTM results are dynamically fused, the load prediction precision is improved, optimal dispatching of the power distribution network is supported, the clean heating operation cost is reduced, and the wind and light abandoning phenomenon is reduced.

The lithium ion battery SOC prediction method based on the CEEMDAN and the optimized BiLSTM comprises the following steps: decomposing preprocessed operation data by adopting adaptive noise complete ensemble empirical mode decomposition CEEMDAN to obtain a plurality of intrinsic mode functions (IMF) and a residual component; constructing a BiLSTM model, wherein the model comprises a forward LSTM unit, a backward LSTM unit and an output layer; a dropout optimization algorithm is introduced into the BiLSTM model, and a combination of a two-parameter sigmoid activation function and a softsign activation function is adopted as a neuron activation function; and inputting the intrinsic mode function IMF and the residual component into the optimized BiLSTM model, optimizing model parameters through training, and predicting the state of charge SOC of the lithium ion battery by using the trained model. Aiming at the problems of weak data nonlinear processing capability, model overfitting, gradient dispersion and the like in the existing SOC estimation method, the invention provides a lithium ion battery SOC prediction method based on CEEMDAN and BiLSTM optimization, so as to improve the SOC prediction precision and model stability.

This invention belongs to the field of image processing and relates to a Retinex-based low-light image enhancement method. It reveals degraded content in the image through a degraded appearance restorer and performs Retinex decomposition using two branches consisting of convolution and a sigmoid activation function to estimate the reflectance and luminance components, respectively. After obtaining the estimated reflectance and luminance components, gamma correction is used to adjust the luminance component. This invention decouples the low-light image into a three-channel color map and a grayscale detail map to maintain consistency with the target image in color and detail representation. Furthermore, this application provides an unsupervised loss function to constrain the solution space of the Retinex decomposition, thereby improving adaptability in unknown and complex scenes. Extensive experiments demonstrate that this method outperforms state-of-the-art methods.

The present invention discloses a remote printer monitoring and fault prediction system based on the Internet of Things, which relates to the technical fields of the Internet of Things and artificial intelligence. It includes a data acquisition module for real-time collecting the device status data of the remote printer and performing preprocessing; a data processing module for integrating the preprocessed device status data into a data matrix and a label vector; a model construction module for using a random forest model and combining with the Gumbel-Sigmoid activation function to output a final feature set; and a fault prediction module for using a bidirectional LSTM to construct a fault prediction model to output the predicted fault probability at the current time point and divide the fault interval data pool. By introducing the combination of the random forest algorithm and the Gumbel-Sigmoid activation function, the feature selection process can be automatically optimized, and the feature set crucial for fault prediction can be accurately screened out. The bidirectional LSTM model is used to capture the time series information, so that the fault prediction accuracy and real-time response ability are significantly improved.

A gas overflow detection method based on contrast learning, comprising the following steps: S1, a contrast learning training stage, a model weight of picture feature extraction is trained, the model comprises three branches respectively for feature extraction, then three features are fused to obtain a final picture feature I o ; S2, the last model in step S1 is frozen, and two fully connected layers are connected thereafter, a nonlinear feature is obtained through a sigmoid activation function between the fully connected layers, finally a feature map of a specific dimension size of the feature is output, an output of positive and negative classification is obtained, representing the classification of whether there is gas overflow; S3, a preset gas leakage confidence threshold value, a picture and a thermal map acquisition device are fixed at a certain angle to fixed-point shoot the device to be acquired, an RGB three-channel picture corresponding to the device and a corresponding single-channel thermal map are acquired, the two fused pictures are input into the model obtained in step S2 for gas overflow classification prediction. Through the present application, gas leakage detection can be better recognized.

The application discloses a bridge risk identification method and system fusing integrated learning and an attention mechanism, and the method comprises the following steps: processing an i-th input feature in an input data set to obtain a reconstructed feature vector; constructing XGBoost and RF-MLP models to perform deep feature extraction, and acquiring fusion features by using an attention mechanism; inputting the fusion features into a back-end multi-task perception decoder, and using parallel decoding branches to decouple and identify the fusion features, wherein a first full connection layer and a first Softmax activation function are used as a risk type decoding branch to output a risk type probability distribution; a second full connection layer and a second Softmax activation function are used as a position decoding branch to output a position probability distribution; a third full connection layer and a Sigmoid activation function are used as a damage degree decoding branch in cooperation with linear mapping to output a damage degree quantitative index; and the application has the advantage of high identification precision.