807 results about "Backpropagation" patented technology

The invention provides an intelligent drilling speed prediction method based on physical feature guidance and multi-source information fusion, and relates to the technical field of intelligent drilling speed prediction, and the method specifically comprises the following steps: collecting multi-source heterogeneous data from a drilling real-time database, a logging system, a logging system and a geological database; constructing a dual-channel deep learning prediction model, wherein the dual-channel deep learning prediction model comprises a dual-channel convolution feature extraction module, a feature fusion module, a time sequence fusion module, a time sequence modeling module and a full connection layer which are connected in sequence; obtaining a predicted drilling speed by using a dual-channel deep learning prediction model; a joint loss function is constructed by considering a data driving error and a physical constraint error, an error is calculated according to the joint loss function, and network parameters are updated through back propagation; carrying out loop iteration training until convergence; and the trained dual-channel deep learning prediction model is used for drilling speed prediction. According to the technical scheme, the problems that in the prior art, a mechanism model is insufficient in precision, and a data driving model is poor in reliability are solved.

Disclosed in the present invention is a lightweight defect detection method based on hybrid multi-scale knowledge distillation. The method comprises: constructing a dataset; constructing a teacher network model and a lightweight student network model; using the dataset to train the teacher network model, and saving a weight file of the trained teacher network model; and loading into the teacher network model the saved weight file of the teacher network model, inputting defect images in the dataset into the teacher network model and the student network model to respectively obtain first multi-scale features and second multi-scale features, respectively inputting the first multi-scale features and the second multi-scale features into a cascaded knowledge blending module to obtain final deeply fused first multi-scale features and final deeply fused second multi-scale features, then calculating a hybrid multi-scale knowledge loss, and in combination with the prediction loss of the student network model, using a backpropagation algorithm to update network parameters, so as to obtain a trained lightweight student network model for implementing defect detection of intelligent manufacturing products. The cognitive ability and recognition performance for defects of different scales are improved.

The invention discloses a cross-source data three-dimensional reconstruction method and system based on improved Gaussian sputtering, and the method comprises the steps: collecting an unmanned plane inclined image and a ground panoramic image of a target region, and constructing a time-space correlation data set; based on multi-view geometric constraints, space-time coding matching point pairs are established through an adaptive feature pyramid, intelligent incremental cross-source data sparse reconstruction is carried out, and point cloud and camera parameters are output; adopting improved Gaussian sputtering, compressing a three-dimensional Gaussian kernel into a two-dimensional Gaussian primitive through double tangent vector constraint, and fitting surface geometry to realize multi-scale reconstruction; and optimizing primitive parameters by using a differentiatable renderer, completing multi-scale fine reconstruction through gradient back propagation, and generating a high-precision three-dimensional model. According to the method, multi-scale accurate geometric prior input and accurate camera poses are provided for three-dimensional reconstruction, the dependence on professional manual operation in a traditional three-dimensional reconstruction method is greatly reduced, and meanwhile, the geometric accuracy and visual fidelity of a reconstruction result are remarkably improved.

A machine learning model may be configured for training using an associated learning technique. A model configured for end-to-end backpropagation may adapted for associated learning by introducing functions for projecting hidden vectors and labels to a shared representation space and for reconstructing labels from representation vectors. An associated learning loss may be calculated at each layer, with the resulting gradients backpropagated locally through that layer rather than all layers. A reconstruction loss may be calculated using each layer's output including the predicted label. Training by associated learning may be parallelized (e.g., layer by layer) to yield efficiency gains. In addition, associated learning training may be more robust to training label errors. The resulting model may be used to, for example, predict data sequences in an autoregressive manner in which subsequent portions of the output data sequence are predicted in part based on previous predicted portions of the output data sequence.

A hybrid neural network architecture is disclosed that integrates matrix multiplication-free (MatMul-free) transformation layers with spiking neural network (SNN) layers for efficient, low-power computation. The system includes an interface module configured to convert intermediate continuous-valued data from MatMul-free layers into a spike-compatible format using encoding techniques such as rate coding, phase coding, or threshold-based conversion. The SNN layers process the spike-encoded data in an event-driven manner, enabling sparse, temporal inference. Training is supported by a hybrid optimization strategy combining backpropagation in MatMul-free components with surrogate gradient descent or spike-timing-dependent plasticity (STDP) in SNN layers. The architecture reduces computational complexity, supports real-time adaptability, and enables deployment in energy-constrained environments such as edge devices and neuromorphic platforms. The system may be implemented in hardware, software, or a co-designed pipeline optimized for dynamic sensor data, control signals, or continuous inference tasks.

To create high-performance metasurface devices (110) in an inverse design process over a large design space (100), a deep neural network may be used as a surrogate model in lieu of a full physics simulation to more efficiently predict figures of merit for given input metasurface topologies during the iterative topology optimization. The neural network may also serve to efficiently compute, via backpropagation, gradients of the figures of merit with respect to design parameters, as are used to update the topology in each iteration.

The invention relates to a textile product defect identification method based on improved YOLOv11. The method comprises the following steps: acquiring a textile product defect image data set; performing pretreatment; dividing into a training set and a verification set; the method comprises the following steps: introducing MConv into a YOLOv11 backbone network, adding a CCIAP module behind a C2PSA module, and applying BiFPN in a path aggregation network; performing prediction through YOLO Head to obtain N prediction feature maps; the overall loss of the network is calculated, and network parameters are optimized through back propagation; predicting the verification set image through a network to output AP values of various categories; repeating the above steps to obtain a trained YOLOv11 network; and detecting the test image or video by using the trained detector to obtain a detection result. According to the method, the MConv is introduced into the YOLOv11 network to enlarge the receptive field, the CCIAP module is added behind the C2PSA to improve the feature extraction capability, and the BiFPN is applied to the Neck layer to enhance the feature fusion capability, so that the target detection precision is improved and the real-time detection of textile product flaws is realized under the condition that the reasoning speed is not influenced.

The invention provides a target layout data generation method, electronic equipment and a readable storage medium. According to the method, the PCB layout generation process is converted into the iterative optimization task of the neural network model, and dual evaluation on the line length and the density distribution in the total loss function is combined, so that the problems of low layout quality, low optimization efficiency and the like in a traditional layout algorithm are effectively solved. The gradient is calculated through back propagation, the positions of the components are updated, it can be ensured that the lengths of the electrical connecting wires and the space uniformity of the components are optimized at the same time in the layout process, and therefore the overall performance and stability of layout are improved. In addition, in combination with a preset training termination condition, overtraining is avoided, and the calculation efficiency is improved. Finally, the generated optimized layout data has higher quality and consistency, and global optimization can be realized in complex design.

The invention relates to an underwater wireless sensor network path sensing routing method based on deep reinforcement learning, which comprises the following steps that: firstly, a node constructs and periodically updates a transmission preference model based on local and neighbor node interaction information; secondly, deploying a deep reinforcement learning model at each underwater sensor node to perform distributed routing strategy learning; and finally, generating a global guide vector by the sink node according to the routing path information of the received data packet, reversely spreading the global guide vector to the source node, fusing the global guide vector with a local transmission preference vector of the node to generate a guide reward, optimizing the deep reinforcement learning model, and updating a routing strategy. According to the method, the problems of difference and complexity of underwater transmission tasks can be solved, and the network data transmission efficiency and the overall service quality are improved in combination with local preference and global guidance while the node online learning is kept to adapt to the dynamic underwater environment.

The invention discloses a battery life prediction method based on Bayesian optimization and a physical information neural network, and belongs to the field of battery health monitoring. The method comprises the following steps: collecting and preprocessing lithium battery charging and discharging data, and then dividing the data into a training set and a test set; then, extracting a dQ / dV curve and other related features, constructing a physical information neural network integrated with physical constraints, and optimizing hyper-parameters of the physical information neural network by using a Bayesian optimization algorithm; and training the model by using a training set, updating parameters through back propagation in the process, evaluating precision by using a test set, and adjusting a strategy. And finally, inputting the characteristics of the lithium battery to be predicted into the trained model, outputting a residual service life prediction value, and comparing with actual data evaluation. According to the invention, reliable lithium battery life prediction can be provided for the photovoltaic energy storage system, and the system operation and maintenance efficiency can be significantly improved, the maintenance cost can be reduced, and the safe and stable operation of the energy storage system can be ensured by early warning the health state of the battery in advance.

The invention discloses an underground water pollution diffusion model construction method based on multi-modal data fusion, and the method comprises the steps: fusing the dynamic time sequence characteristics of a monitoring well and the static space attributes of a geological field, and generating an initial state vector with rich information for a graph node; based on the instantaneous water level difference between the nodes and the equivalent permeability coefficient, a dynamic graph topological structure which evolves along with time and represents the hydraulic connection relation is constructed; deducing a future state through numerical integration by using a graph neural network model with a node state derivative as output; and constructing a composite loss function containing a data fitting item and a physical law residual item, and training the model through back propagation. According to the method, the dynamic graph of physical driving is constructed and the physical equation constraint is introduced, so that the model prediction has the data driving precision and the reliability of the physical mechanism, and the generalization ability and the prediction precision of the model under the complex hydrogeological condition are improved.

The invention relates to the technical field of data fusion, in particular to a multi-source electrocardiosignal correction method and system based on adaptive fusion, and the method comprises the following steps: constructing a multi-channel input tensor, extracting local features through a weight calculation network, carrying out the adaptive weight fusion and dimension reduction of multiple paths of signals, and carrying out the correction of the multi-source electrocardiosignal. A nonlinear mapping relation is established through a deep reconstruction network, a standard waveform is reconstructed, and network parameters are optimized based on reconstruction error reverse iteration. According to the method, local neighborhood features of multichannel signals are extracted by constructing a weight calculation network, a dynamic channel weight sequence reflecting the real-time contribution degree of a signal source is constructed, the amplitude intensity is adaptively adjusted according to the signal quality, unstable channel noise interference is effectively inhibited, and high-quality signal components are enhanced; a deep reconstruction network is used for carrying out nonlinear feature transformation on a fusion sequence, accurate mapping from non-standard input to standard lead waveforms is established, and weight distribution and optimization of signal reconstruction parameters are achieved in combination with an error back propagation mechanism.

The invention provides a wind power prediction method based on a frequency domain space-time diagram convolutional network. The method comprises the following steps: firstly, respectively constructing a spatial proximity weight matrix and a similarity matrix based on a fan position and historical feature data, and adaptively fusing the spatial proximity weight matrix and the similarity matrix into an adjacent matrix; performing high-dimensional embedding on the normalized features, and converting a time domain signal to a graph frequency domain by using graph Fourier transform GFT; designing a prediction and reconstruction double-branch network, extracting key frequency features through discrete Fourier transform (DFT), causal convolution and channel attention, and performing inverse transformation reduction; and finally, constructing a joint loss function of prediction and reconstruction loss weighting, and realizing end-to-end training through back propagation and an Adam optimizer. According to the method, spatial topology and time-frequency dynamic characteristics are fused, prediction and reconstruction advantages are combined, and the precision and robustness of wind power prediction are improved.

The invention discloses a large-scale building group layout optimization method based on deep learning. The method comprises the following steps: S1, obtaining a three-dimensional rasterized matrix and a meteorological environment vector of a to-be-optimized building group; s2, on the basis of a pre-trained three-dimensional building layout-wind field mapping network, generating a global wind environment distribution map according to the three-dimensional rasterized matrix and the meteorological environment vector; s3, based on the global wind environment distribution diagram, calculating a building layout parameter gradient through a gradient back propagation optimizer, and generating a layout optimization scheme; and S4, based on the layout optimization scheme, outputting optimized building group space layout parameters. Intelligent optimization of large-scale building group layout is realized through a deep learning technology, and urban wind environment quality and planning and design efficiency are remarkably improved.

The invention relates to the technical field of semiconductor device preparation, in particular to a super-junction MOSFET preparation process parameter optimization method based on simulation analysis, which comprises the steps of generating a simulation model according to a preparation process, executing simulation based on a plurality of groups of initial process parameter sets and the simulation model, performing simulation electrical performance test on a plurality of simulated super-junction MOSFETs, and performing simulation analysis on the simulated super-junction MOSFETs. Performing mapping relation learning on the plurality of groups of initial process parameter sets and the plurality of electrical performance data sets by utilizing a pre-constructed back propagation agent model, performing optimization solution on a mapping relation model, performing decision screening on an optimal solution set, and importing an optimal process parameter combination into a pre-constructed TCAD simulation model for verification to obtain a TCAD simulation model; and completing super junction MOSFET preparation process parameter optimization based on simulation analysis. According to the method, accurate estimation of the optimal process parameter combination is realized through simulation analysis, and dynamic correction is carried out when the optimal process parameter combination does not reach the expectation, so that the super-junction MOSFET process parameter combination which is better in performance and feasible to prepare is obtained.

The invention relates to a cross-domain few-sample fault diagnosis method based on a geometric perception coordinated penalty prototype network, and belongs to the technical field of fault diagnosis, and the method comprises the following steps: S1, dividing samples through employing vibration monitoring signals in a data set; s2, obtaining respective support sets and query sets through source domain samples and target domain samples, and generating meta-tasks used in training and testing stages; s3, extracting input sample features by using a feature extractor; s4, mining an internal geometric structure of the data by using an adaptive manifold mapping module (AMMM), and converting sample features to manifolds; s5, calculating to obtain a coordination penalty prototype HPP; s6, classification is carried out through a dynamic high-sensitivity distance DHSD, loss is calculated, model parameters are updated through back propagation, and training of the GAHPPN is completed; and S7, quickly adapting to the meta-task of the target domain through the trained GAHPPN model, and diagnosing the health state of the target domain sample.

The invention discloses a private data security sharing method based on federal learning, particularly relates to the field of private data security protection and sharing, and is used for solving the problems of insufficient model credibility and lack of precision control of data exchange in the existing cross-mechanism data collaboration process. According to the method, homomorphic encryption processing is carried out on local data of all participants, a ciphertext verifiable calculation task is constructed in combination with federated learning, forward propagation and back propagation calculation of a ciphertext are carried out by outsourcing calculation nodes, credible verification is carried out on a calculation process by utilizing zero-knowledge proof, and gradient aggregation and model updating are completed in a ciphertext domain; after model training is completed, based on federal learning model output, intelligent judgment is conducted on the sharing value of local data samples, ciphertext re-encryption and data exchange control are driven, multi-party safe and controllable data circulation is achieved, and therefore on the premise that privacy safety and compliance requirements are guaranteed, cross-mechanism data collaboration efficiency and decision-making precision are improved.

The invention discloses a physical object three-dimensional positioning method and system based on video data, and relates to the technical field of three-dimensional positioning. The method is used for solving the problems of low object space positioning precision, unstable pose estimation and poor time sequence continuity in a video scene. Firstly, an input video stream is analyzed, an object segmentation mask is generated, feature points are extracted, camera motion parameters are calculated through inter-frame matching, and a scene sparse three-dimensional point cloud is reconstructed; a candidate three-dimensional bounding box is generated according to the segmentation mask and the point cloud, an optimal bounding box is selected by combining geometric matching degree and feature similarity evaluation, and a preliminary three-dimensional positioning result of the object is obtained; a positioning result and historical frame motion data are fused, a space-time constraint optimization model is constructed, and the six-degree-of-freedom pose of the object is solved; and finally, neural radiation field representation is established based on the pose, the pose and neural parameters are optimized through combination of micro rendering and back propagation, continuous and accurate updating of the three-dimensional position of the object is realized, and the positioning stability and robustness in a dynamic scene are improved.

The embodiment of the invention provides a GNN-based financial fraud detection method, and belongs to the technical field of data processing, and the method specifically comprises the steps: constructing a financial heterogeneous transaction graph, and carrying out the normalization; for each relation type, calculating a PPR matrix of the relation type, and calculating a multi-scale prior weight vector for each node based on the matrix; inputting the multi-scale prior weight vector and the node features into a multilayer graph neural network model, and generating multi-scale fusion representation of each node under each relation type; generating an output representation of the node in the current network layer based on a relation attention mechanism; generating a final feature representation of each node based on a hierarchical attention mechanism; inputting the final feature representation into a classifier to obtain a probability prediction value that each node is an abnormal node; and calculating a loss function by using real data with labels, and optimizing parameters of the multi-layer graph neural network model and the classifier through back propagation. Through the scheme of the invention, the detection accuracy and robustness are improved.

The invention relates to the technical field of computer vision, and discloses a Monte Carlo tree search-based camera lens assembly focusing calibration algorithm and system, and the algorithm comprises the steps: constructing a three-dimensional search space comprising a focus position, a horizontal inclination angle and a vertical inclination angle, and combining a high-contrast cross wire test pattern to generate a local definition matrix; calculating an initial inclination angle based on the matrix to construct a root node of the Monte Carlo tree; according to the algorithm, an optimal search path is dynamically selected by applying an upper confidence bound strategy, child nodes are expanded in a three-dimensional space, physical adjustment is executed, and optimal three-dimensional configuration parameters are finally determined based on node scores through back propagation iterative optimization. The method breaks through a traditional one-dimensional search framework, effectively avoids a pseudo peak trap caused by inclination of a lens-sensor axis, improves the calibration precision and efficiency of the full-view-field definition of the camera, and achieves the precise positioning of a global optimal solution under a complex assembly tolerance.

The invention discloses an employee ability assessment method and system based on a structure attribution neural network, and a medium, and belongs to the field of power systems, and the method comprises the steps: extracting the partial derivative information of an ability assessment model based on a gradient back propagation mechanism, determining the gradient sensitivity based on each partial derivative information, and determining a weight parameter; processing the real-time evaluation event according to the weight parameter to obtain a target evaluation event, and performing matrix processing and division according to the index service dimension to obtain a plurality of sub-matrixes, so that the capability evaluation model splices feature extraction results output after extraction of the plurality of structure sub-networks for each sub-matrix into an intermediate vector, the intermediate vector is predicted through the summary regression network, a target evaluation result is output, the structure dimension of each structure sub-network is in one-to-one correspondence with the service dimension, and the model adopts a loss function based on attribution guidance, so that objective and accurate evaluation of the ability of the to-be-evaluated employee can be realized through implementation of the method and the system.

The invention discloses a KAN adaptive function combination method and device based on multi-scale sparse branches, and belongs to the technical field of artificial intelligence and machine learning. The method comprises the following steps: starting a dynamic selection mechanism and initializing model parameters; performing standardization processing on the input data, and calculating a function selection weight through multi-scale feature extraction and a hierarchical attention mechanism; a correlation function module is screened and activated based on the weight, and a sparse calculation path is formed; executing feature calculation of the activation module and integrating and outputting by adopting a weighted fusion strategy; iteratively executing module selection and feature fusion in the deep network, and constructing deep feature representation; and finally, a prediction result is generated based on the fusion features, and model optimization is realized through loss calculation and back propagation. According to the method, the reasoning speed and the calculation efficiency are remarkably improved, good input adaptability is achieved, energy consumption can be reduced, the model interpretability can be enhanced, and the method is suitable for application scenes such as computer vision, natural language processing and industrial control.

A rock porosity prediction method based on small-sample while-drilling data inversion is provided, including: collecting while-drilling parameters during laboratory or field drilling processes, including parameters such as torque M, propulsion F, rotational speed N, drilling speed V, drill pipe amplitude A, and vibration acceleration a; inputting the different types of multi-stage denoised while-drilling parameter data into a backpropagation regression-genetic algorithm (BP-GA) model for discrete point elimination, data augmentation, and iterative calculation to form a new while-drilling dataset; inputting the time-frequency domain feature graphs of while-drilling parameters for rocks with different porosities into a rock porosity convolutional neural network (VG-CNN) prediction model, obtaining an inversion model between the time-frequency domain features of while-drilling parameters and porosity through training and learning, ultimately achieving inversion prediction of rock porosity according to real-time field while-drilling parameters.

The invention discloses a peanut variety identification method and system based on machine vision, and the method comprises the steps: obtaining original image data of a peanut sample, and obtaining an enhanced peanut image set; obtaining a quantized shape description vector from the enhanced peanut image set, synthesizing a texture description vector and a color quantization vector to obtain a fused multi-dimensional feature vector, constructing a training data set based on the fused multi-dimensional feature vector, extracting a deep mode by adopting a convolutional neural network architecture, and optimizing weight parameters through back propagation to obtain a deep mode; obtaining a peanut variety identification model; and inputting the fused multi-dimensional feature vector corresponding to the tested peanut image into the peanut variety identification model to obtain a final peanut variety identification result. According to the invention, through combination of multi-dimensional feature integration and deep learning, the accuracy and stability of peanut variety identification are significantly improved.

The invention relates to the field of computer vision and remote sensing image processing, and discloses a semi-supervised remote sensing target detection method, system, medium and equipment based on spatial resolution guidance, and the method comprises the steps: maintaining all types of GSD Gaussian distribution parameters in labeled and unlabeled data under a teacher-student framework; calculating category-level weights of the unlabeled images according to the matching degree of GSD distribution of the unlabeled images and GSD distribution of pseudo label categories in labeled data; calculating an image level weight based on the GSD distribution difference of the pseudo-label category between the labeled data and the unlabeled data; combining the supervised loss of the labeled data with the unsupervised loss of the double-weighted unlabeled data to serve as total loss; updating the student model through back propagation, and updating the teacher model through index moving average; and performing target detection by using the trained model. According to the method, pseudo label noise caused by resolution difference can be reduced, and stable training and detection performance improvement can be realized under a low labeling rate.

The invention discloses a helicopter prediction control method based on a physical mechanism neural network and a helicopter, and the method comprises the steps: building a helicopter prediction model, so as to describe a relation between a future state of the helicopter and a current state quantity and a control variable; constructing a predictive control model based on a physical mechanism neural network; generating a control quantity sequence by adopting a forward reasoning process of the neural network; according to the control quantity sequence and the state quantity, adopting the helicopter prediction model to perform multi-step iterative evolution in an autoregression form to obtain a predicted helicopter state trajectory; calculating a loss value by adopting a loss function of an embedded physical mechanism according to the expected instruction and the predicted helicopter state trajectory, and performing back propagation on the neural network to update neural network parameters; and inputting the state quantity obtained in real time and an expected instruction into the trained prediction control model to obtain an optimal control sequence, and applying the first control quantity to realize rolling time domain closed-loop control.

The invention provides a Bayesian network granularity prediction method and system based on physical constraints. The method comprises the following steps: collecting historical process parameters and corresponding granularity distribution indexes D10, D50 and D90 of a ternary hydroxide synthesis process; preprocessing the process data to generate an input feature matrix; a Bayesian attention neural network model is constructed, probability distribution and confidence intervals of particle size distribution parameters are output through Bayesian back propagation training, and model output layers correspond to D10, D50 and D90; and embedding physical constraint terms such as range, distribution width, proportion and dynamic process and / or PBE constraint terms in the loss function, and balancing prediction precision and physical rationality and quantifying uncertainty based on the comprehensive loss function. The system comprises a data acquisition module, a data processing module, a model construction module and a prediction module. According to the method, process data, physical constraint and data driving are fused, the accuracy, real-time performance and adaptability of particle size distribution prediction are improved, and intelligent manufacturing and process optimization of the ternary hydroxide precursor are supported.

The invention discloses a multiphase flow real-time flow pattern recognition method based on updip pipe pressure and a lightweight deep neural network model, adopts updip pipe pressure data, designs a training strategy, proposes a YI-Net network model, belongs to the field of deep learning, and is used for flow pattern recognition. The method comprises the steps of data preprocessing, model reasoning, loss calculation, back propagation optimization of model parameters and the like. In the data preprocessing process of the training stage, the generalization ability of the model is improved through a data enhancement training strategy, such as Gaussian noise addition, amplitude scaling, overall time translation and the like. The YI-Net model comprises two parts of a backbone network (backbone) and a head (head), and various one-dimensional calculation operations such as one-dimensional convolution and the like are adopted. And the backbone is used for extracting the characteristics of the flow pattern sample. And the head is a Classify layer and is used for determining the category of the backbone output characteristics. The flow pattern recognition scheme provided by the invention has the characteristics of good real-time performance and small parameter quantity, and provides a solution for the field of flow pattern recognition.

The invention relates to the technical field of artificial intelligence, in particular to a risk assessment model construction method based on multi-modal data fusion and a neural network, and the method comprises the following steps: collecting videos, detecting clothing colors, extracting audio frequency domain features, counting text emotion keywords, and constructing a multi-modal risk feature set; the method comprises the following steps: generating risk transmission weight data, mapping the risk transmission weight data to a causal matrix, carrying out back propagation correction to generate risk transmission weight data, judging conflicts according to weight change to generate risk conflict identification information, reconstructing a causal chain, extracting key factors to construct a risk propagation path network, inputting the risk propagation path network into LSTM (Long Short Term Memory) assessment, and carrying out classification to generate a multi-modal fusion risk assessment model. Composite features are constructed through multi-source signals, an adjustable conduction relation is formed in combination with weight changes, self-correction is triggered according to node symbol differences to clear conflict components to highlight key factors, conduction paths are generated according to an intensity descending order, and evolution expression is formed in time sequence modeling. The model is stable, interpretable and practical guiding significance in cultural heritage protection and cultural scene risk management and control.

The invention relates to an oil and gas pipeline magnetic flux leakage detection defect quantification method based on a composite convolutional neural network. The method comprises the following steps: 1) a data acquisition stage: carrying out saturation magnetization detection on a pipeline by adopting a pipeline internal detector to obtain magnetic flux leakage internal detection data containing defects; 2) a data preprocessing stage: performing normalization processing on the magnetic flux leakage data, and constructing a defect quantification data set; 3) a model construction stage: constructing a composite convolutional neural network model; 4) a model training stage: inputting a defect quantification data set sample into the model for training; and 5) a model test stage: performing feedback and iterative optimization for multiple times through a back propagation optimization algorithm of the model, finally outputting optimal model parameters in the training process, and exporting and storing the model parameters. Based on the pipeline magnetic flux leakage detection data, high-precision quantification of pipeline defect size parameters can be realized, and the accuracy and reliability of defect evaluation can be improved.