914 results about "Network computing" patented technology

The invention relates to the technical field of artificial intelligence and medical diagnosis, in particular to a disease diagnosis prediction method and system based on a graph neural network. The disease diagnosis prediction method based on the graph neural network comprises the five steps of heterogeneous medical knowledge graph construction, adaptive node embedding representation, hierarchical graph attention network modeling, incremental learning dynamic graph updating and multi-dimensional feature input and result output. The invention discloses a disease diagnosis and prediction system based on a graph neural network. The system comprises a multi-source data acquisition module, a heterogeneous graph construction module, a self-adaptive embedding module, a graph network calculation engine, a dynamic updating module, a disease prediction module and a feedback optimization module. According to the method, the multi-modal heterogeneous knowledge graph is constructed to integrate the multi-dimensional data of the patient, and the hierarchical graph attention network and the dynamic incremental learning are combined, so that the accurate prediction of the disease risk and the visual explanation of the pathological association path are realized.

The invention relates to the technical field of data processing, in particular to an intelligent analysis system for electric energy quality and read data, which comprises a data fusion module, a modeling module, an analysis module, a detection module, an optimization module and a control module. The data fusion module aligns data of an electric meter terminal, power grid monitoring equipment and an environment sensor through a sliding time window, and the analysis module calculates a power grid node loss transfer coefficient based on a graph neural network and fuses transformer no-load loss and line contact resistance parameters to generate a dynamic line loss evaluation matrix. And the detection module identifies the power consumption characteristic deviation degree through a random forest classifier, and generates a priority management strategy in combination with a multi-dimensional abnormal scoring model. The control module adaptively selects a communication protocol to execute a regulation and control instruction according to a network state, a heartbeat detection mechanism feeds back operation data of a governance device in real time, model parameters are driven to be iteratively updated, and dynamic cooperation of power supply quality optimization and line loss governance is achieved.

The invention relates to a remote sensing target detection method and device and a medium, and the method comprises the steps: inputting a feature map to a backbone network for feature extraction, inputting an extracted feature tensor into a multi-branch expansion convolution structure, and extracting multi-scale features through convolution kernels with different expansion rates. Then, multi-scale features are fused through a space and channel double-path attention mechanism, and enhanced features are generated; and the enhanced features are further input into a cascade pooling module to generate multi-level reconstruction features, and weight coefficients are calculated through a gating fusion network to carry out weighted fusion, so that multi-scale fusion features are obtained. Next, these features are input into an asymmetric decomposition convolutional layer for downsampling, and dynamic channel attention calibration is performed to generate channel enhanced features. And finally, inputting the feature map processed by the backbone network and the neck network into a detection head network, and outputting a target bounding box and category prediction. According to the method, high-precision detection of multi-scale rotating targets and high-density small targets is realized in a complex remote sensing scene.

The invention relates to the technical field of environment detection, and discloses a GIS map-based multi-class environment parameter automatic prediction method, which comprises the following steps: collecting noise intensity, vibration spectrum, sewage turbidity, illumination intensity and PM2.5 concentration in real time through a distributed sensor network, combining social behavior data, carrying out space-time alignment, and generating a multi-dimensional space-time matrix; calculating an energy overlapping degree by adopting a space-time diagram attention network, marking a high-risk collaborative pollution area, and predicting a pollution diffusion path; constructing a self-adaptive prediction model containing physical and behavior driving channels, and dynamically adjusting the weight to optimize the prediction precision; based on the optimization model, reversely deducing a pollution source of an overproof area, matching equipment characteristics and generating a control instruction; and opening an AR (Augmented Reality) interface to verify the treatment effect, and when the virtual-real data deviation exceeds a data deviation threshold, triggering federal learning to update the global model, and generating an environmental protection compliance report. According to the invention, the accurate decision-making efficiency of environmental governance can be improved.

The invention relates to the technical field of information resource fusion management, and particularly discloses a fusion and management system for multi-source heterogeneous science and technology information resources. Analyzing an equipment fault chain from an unstructured text of a historical operation and maintenance log, extracting a rated parameter constraint from a structured table of an equipment manual, and collecting an operation feature vector from a real-time sensing data stream to generate a knowledge graph containing N entity relationships; based on an entity attribute constraint rule of the knowledge graph, designing a bidirectional attention mapping network to calculate semantic similarity weights of multi-source data and knowledge nodes, and generating a graph embedding vector set with weight marks through Hadamard product operation; according to the method, the embedded vector set is input into the pre-trained graph neural network model, and the root cause equipment set causing feature offset is positioned, so that efficient fault diagnosis and positioning are realized, decision support is provided for a subsequent preventive maintenance strategy, and the reliability and the operation and maintenance efficiency of the system are improved.

The invention discloses a cloud platform auditing method based on multi-modal data processing, and relates to the technical field of cloud platform auditing. Independent features of text, image, audio, video and log data are extracted, each piece of modal information can be subjected to refined analysis in a corresponding feature space, and in a cross-modal feature fusion stage, a multi-modal feature fusion model is obtained. The correlation between different modal data is calculated by using a cross-modal consistency comparison network CMCN, and feature alignment is optimized through an attention mechanism, so that multi-modal information can be mutually verified, and misjudgment and missed judgment caused by independent auditing are reduced; in the auditing decision-making link, an auditing model based on deep learning is constructed, and end-to-end illegal content judgment is realized in combination with anomaly detection and a self-adaptive risk control threshold value, so that the intelligent level of an auditing system is improved; in addition, the suspicious areas of the image and the video are visually marked, and an auditing conclusion is generated by utilizing a natural language to generate an NLG model, so that the auditing result is more transparent.

The invention provides a scene adaptive projection vehicle lamp system based on deep reinforcement learning and a control method, and relates to the technical field of intelligent vehicle lamps and automatic driving perception systems. Comprising a multi-modal sensing module, a feature fusion module, a strategy generation module and an execution module. The multi-mode sensing module is used for collecting environment state data and performing primary processing to form an environment data information flow; the feature fusion module is used for generating a unified environment feature vector for the environment data information flow; the strategy generation module is used for receiving the environment feature vector and generating a vehicle lamp adjustment strategy through multi-layer neural network calculation; evaluating a result obtained by executing the vehicle lamp adjustment strategy based on the vehicle lamp, and optimizing strategy parameters of the strategy network based on a PPO algorithm; the execution module is used for controlling the vehicle lamp according to the vehicle lamp adjustment strategy output by the strategy network. The intelligent level of the vehicle lamp is remarkably improved, and the system is widely applied to night driving assistance, urban interaction prompt and low-visibility driving scenes.

The invention discloses an electric power analysis method and system based on artificial intelligence, and relates to the technical field of electric power system intelligent analysis. Establishing a hierarchical depth map neural network model, and setting a cross-layer information interaction channel to connect each sub-network; executing a bidirectional knowledge distillation algorithm, extracting a rule set from neural network output, and forming a parameter-rule bidirectional mapping matrix; deploying a federal reinforcement learning architecture, performing parameter aggregation on the local models on the distributed power nodes, and generating a power system state evaluation result and a risk coefficient matrix; and starting the self-repairing intelligent agent network, calculating a regulation and control parameter set of the multilevel power system, and outputting a regulation and control instruction. According to the invention, through the hierarchical depth map neural network and a cross-layer information interaction mechanism, collaborative optimization among all levels of the power system is realized, and the operation efficiency and stability of the system are improved.

The invention discloses an event early warning efficient management method and system based on park management, and relates to the technical field of safety management, and the method comprises the steps: constructing a hexagonal cellular unit layout, generating an equipment registry and a network topology structure, and forming a structured data set through data alignment and sliding window verification; the edge server loads an initial weight, calculates a dynamic weight matrix based on the LSTM network, and generates a high-order feature vector set in combination with the graph neural network; inputting the high-order feature vector set into a space-time prediction model to generate an environment evolution trend prediction value, fusing network topology and BIM model parameters to calculate a dynamic risk index, and triggering graded early warning; the command center analyzes the early warning signal to generate a fusion visual interface, historical cases are matched through federal learning, a resource allocation scheme is optimized, and an optimal disposal strategy is output through Monte Carlo simulation. By dynamically adjusting the weight of the sensor, the weight of the infrared sensor in the high-temperature-sensitive area is increased, and the abnormality capturing efficiency is improved.

The invention relates to the technical field of data processing, and discloses a project cost prediction method and system based on a large model, and a storage medium. The method comprises the following steps: standardizing historical cost data to construct a dynamic cost feature knowledge base; inputting the knowledge base into a Transform pre-training large model, and performing fine tuning to obtain a predicted value; obtaining an adjustment coefficient matrix through item similarity clustering and error learning; the design parameters are matched with a knowledge base, and a hierarchical cost table is calculated by using a heterogeneous graph network; and performing time correction on the cost table based on the adjustment coefficient to obtain a prediction result. According to the method, accurate prediction and dynamic adjustment of the project cost are realized, the problem of insufficient element association expression in traditional cost prediction is solved, targeted correction can be carried out according to time and project features, and the prediction accuracy and adaptability are remarkably improved.

The invention discloses a wire harness manufacturing management system and method based on artificial intelligence, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: inputting a structured feature vector and a process constraint condition into a multi-target reinforcement learning model, and generating a process parameter instruction; the process parameter instruction is executed, the product yield is monitored in real time, and cross-process quality tracing is carried out on the abnormal product yield; historical data of abnormal batches in crimping, assembling and testing procedures are traced, the defect root cause probability of each procedure is calculated based on a Bayesian causal network, and primary and secondary root causes of quality defects are obtained and analyzed; and dynamically updating process constraint conditions according to primary and secondary root cause analysis results, adjusting a reward function of the multi-target reinforcement learning model, and re-optimizing a process parameter instruction. According to the invention, the operation state of the production equipment is accurately captured, the production process is optimized and adjusted in real time, and a closed-loop management process is formed, so that the production flexibility, efficiency and product quality are remarkably improved.

The present disclosure relates to a multi-modal brain network computation method associated with structural function, apparatus, device, and medium. The method is applied to train a brain disease prediction model, and the brain disease prediction model includes an association perception dual-channel generation module, a disease feature regression module, a topological structure discriminator, and a time-space joint discriminator. In a model training process, by performing a multi-level interactive fusion learning on a high-order topological feature of brain functional magnetic resonance data and magnetic resonance diffusion tensor imaging data, a multi-modal time series activity signal of each brain region is obtained.

The invention discloses an intelligent operation abnormity monitoring and positioning method for a polypropylene cable, and relates to the technical field of intelligent operation and maintenance of a power system, and the method comprises the steps: collecting multi-source physical signals in the operation process of the cable, and constructing a multi-dimensional feature matrix fusing multiple physical quantities through multi-scale time window division and space mapping processing; extracting space-time coupling characteristics among nodes by using a graph attention embedding network, and training a running state recognition model in combination with a label perception contrast learning mechanism; constructing a cable topological graph based on an identification result, introducing an improved Bayesian space reasoning network, and calculating abnormal probability distribution of each node; a weighted abnormal heat map is further generated, an abnormal propagation path is extracted through an abnormal state flow model and a directional propagation scoring algorithm, and abnormal node positioning and trend evolution prediction are achieved; the method has the advantages of high spatial resolution, high identification precision and good online adaptability, and is suitable for intelligent state perception and abnormity early warning of the polypropylene cable in a complex operation environment.

The invention discloses a traffic protection facility state detection method and system based on digital twinning, and relates to the field of data processing methods for prediction purposes. Constructing a characteristic parameter matrix in the virtual model layer; performing adaptive decomposition on the characteristic parameter matrix to obtain a parameter subspace; establishing a parameter subspace index chain; analyzing the change correlation degree among the state parameters in each parameter subspace, and constructing a correlation degree network; calculating correlation evolution characteristics of the parameter subspace to obtain an evolution characteristic sequence; identifying an associated abnormal region, and positioning an abnormal source; performing state prediction simulation on the virtual model layer to obtain a state prediction simulation result; and predicting a parameter change curve, and when any value in the parameter change curve exceeds a preset safety threshold, outputting an early warning signal to the physical entity layer through the data interaction layer. The method and the device are used for improving the accuracy of traffic protection facility state detection, so that the preventive maintenance effect of the traffic protection facility is improved.

The invention provides a power grid cable loss analysis method and device and a storage medium, and the method comprises the steps: obtaining cable topology data, extracting parameter characteristics, and generating a parameter database; performing matrix conversion based on the parameter database to establish a network calculation model, executing load flow calculation, and outputting voltage power distribution data; establishing a temperature and loss coupling relation model according to the voltage power distribution data, calculating cable temperature distribution, and generating time-varying loss data; and performing multi-dimensional feature analysis on the time-varying loss data, calculating a loss risk index, identifying loss hotspots and generating an optimization scheme. According to the method, the dynamic relation model of the cable parameters and the loss is established, and the cable shielding layer characteristics and the temperature change factors are combined, so that accurate analysis of the power grid cable loss is realized, technical support is provided for identifying loss hotspots and formulating a targeted optimization scheme, and the power grid loss calculation deviation is effectively reduced.

The invention discloses a full-stack NPU system supporting a multistage fault mitigation mechanism, and the system comprises a plurality of NPU functional blocks which are used for executing a neural network calculation task; the plurality of hardware protection modules are used for providing error detection and correction capabilities for at least one NPU functional block in the NPU or a data path in the NPU; and a control mechanism for coordinating operation of the plurality of hardware protection modules. By means of the scheme, NPU functional block calculation errors or data path transmission errors caused by hardware faults and the like can be detected and corrected in time, wrong calculation results are prevented from being continuously used or output, and the accuracy of final results is guaranteed. Meanwhile, even if part of hardware breaks down, an error correction mechanism can also cover errors to a large extent, normal operation of the system is maintained, the overall availability and task success rate of the system are improved, and therefore the reliability and robustness of the NPU are remarkably improved.

The invention discloses a power transmission line icing galloping risk early warning system and early warning method, and belongs to the technical field of icing galloping risk early warning, and the system comprises the following modules: a historical data processing module reconstructs historical monitoring data by using a long and short term memory auto-encoder, recognizes abnormal data through reconstruction errors, and sends the abnormal data to an early warning module; removing outliers based on a 3sigma-dynamic threshold algorithm, and then marking a space-time credibility weight; the terrain compensation module constructs a micro-terrain feature vector, calculates the similarity through a Siamese network, and compensates low-confidence data; the real-time data processing module obtains the icing thickness according to the monitoring data of the current time point and the previous time point; the data acquisition and analysis module fuses historical and real-time data and analyzes a galloping state; and the risk early warning module takes the historical data with the weight and the galloping state information as input, outputs a risk value through the prediction model, and triggers early warning. According to the system, through cooperative work of all the modules, the icing galloping risk is accurately warned in real time, and the safety and reliability of power grid operation are improved.

The invention belongs to the technical field of power system optimization scheduling, and provides a source network load storage intelligent collaborative optimization method, which comprises the following steps of: deploying sensors at four ends of a source network load storage respectively, collecting in real time by utilizing a cloud data center, enabling data of the four ends to be consistent in time sequence through a PTP protocol, constructing a topological graph according to parameters and data, and establishing a source network load storage intelligent collaborative optimization system. Selecting a model in a digital twinning environment for simulation; dividing independent agents at four ends of a source network load storage, setting observation data, an execution space and excitation feedback, forming an excitation item by economy, stability and environmental protection, interactively circulating actual data, a prediction instruction and an excitation value, recording into a sequence, inputting the sequence into a strategy network, and calculating and outputting logarithmic probability gradient to update the parameters of the strategy network; and the intelligent agent completes interactive circulation according to the strategy network, generates a local scheduling instruction, aggregates the instruction to perform weighted calculation, generates a global scheduling scheme, issues the global scheduling scheme to execution equipment, updates parameters by using an average deviation calculated by a deviation vector, resolves the global scheduling scheme and issues the global scheduling scheme to form a closed-loop mechanism.

The invention discloses an online efficient load shedding control method based on a forecast wind field. The method comprises the following steps: firstly, acquiring a load attack angle and a sideslip angle which are calculated in real time on a rocket; amplitude limiting processing is carried out on the load attack angle and the sideslip angle, an attack angle control instruction is generated after gain control network calculation is carried out, and an attitude control system carries out direct attack angle control in a load sensitive area in combination with the load shedding requirement and the guidance deviation correction capacity; and pitching and yaw feedback control instructions are generated, active load shedding control is implemented, and errors generated by attack angle control are compensated. The on-line active and passive load shedding mode covers the real deviation or interference influence during flight, the generated load shedding instruction is better matched with the actual flight, the universality is good, the implementation is easy, and the load shedding effect is better.

The invention discloses a remote medical interaction method based on virtual reality, and particularly relates to the field of remote medical interaction.The method comprises the steps that by synchronously collecting multi-modal tactile data and transmission state parameters of a patient end, a joint feature vector is established, and the transmission delay of each modal tactile signal is predicted in real time. The method comprises the following steps: extracting sensitivity indexes of an operation stage and a tactile signal through a doctor end real-time image, calculating a reward value in combination with a double-branch regression network, updating a modal transmission strategy, and generating a dynamic priority distribution matrix; based on the matrix, optimizing a bandwidth slicing strategy and performing signal compensation by combining a real-time network state and delay prediction; in addition, switching and pathological mutation of surgical instruments are detected, priority distribution is dynamically adjusted, and efficient transmission and operation feedback of tactile signals are ensured.

The invention discloses a mine rescue training and danger dynamic simulation method based on virtual reality, and the method comprises the following steps: S1, modeling a rescue process into a directed graph structure, and outputting a task state diagram bound with a virtual reality scene; s2, collecting an operation behavior, a task state, an environmental parameter and a physiological response, and constructing a training state vector; s3, inputting the state vector into the deep Q network to calculate a jump node evaluation value, and outputting a jump node; s4, modeling the control rule as a rule node, constructing a directed acyclic graph, binding a weight, and outputting a rule graph; s5, performing self-evolution on the rule atlas according to training feedback, and outputting an evolution structure; s6, reasoning in the evolution rule map, fusing the rule recommendation node and the jump node, and outputting a final task node; and S7, executing jump control according to the final node, loading a virtual scene, and pushing environment parameters, task contents and risk information. According to the invention, intelligent path decision and training process adaptive optimization of the mine rescue task are realized.

The invention provides an underwater acoustic target recognition method based on autonomous task perception, and belongs to the field of underwater acoustic signal processing and artificial intelligence. S3, performing feature extraction on the time-frequency spectrogram by the task type extraction network to obtain a task embedding vector, calculating a similarity score, when the maximum similarity score is greater than a set threshold value, outputting a task representation vector, and entering S3, otherwise, inputting features output by the last Transform layer into a classifier; s3, selecting a router according to the task representation vector, selecting a trained expert network by the router, calculating a door control weight, processing universal acoustic features in parallel by the expert network, fusing output of the expert network, fusing deep features and fused features to obtain enhanced features, and enabling the enhanced features output by the last Transform layer to enter a classifier; the invention further provides a system. The problems that the task category autonomous recognition capability is insufficient and the correlation between tasks is ignored are solved.

The invention discloses an enterprise credit risk assessment method based on big data collection, and relates to the technical field of big data analysis, and the method comprises the steps: constructing a space-time fusion engine, carrying out the fusion through combining a cross-modal attention mechanism, generating a space-time fusion feature, and carrying out the adversarial training through a gradient inversion layer and a domain classifier, eliminating space-time fusion feature distribution differences; based on space-time fusion features, constructing a causal graph skeleton by adopting a conditional independent test algorithm, quantifying causal effect intensity among nodes through a machine learning model, constructing a risk conduction dynamic model, and predicting risk conduction intensity based on the causal effect intensity; and based on the risk conduction intensity, calculating a risk index weight through a dynamic game network, generating an enterprise risk score in combination with the space-time fusion feature, and generating a dynamic risk score through a time sequence neural network. According to the invention, by constructing the space-time fusion engine and combining the cross-modal attention mechanism, efficient fusion of multi-modal data is realized, and the accuracy and robustness of feature expression are improved.

The invention discloses a resource scheduling optimization method and system based on deep learning, and particularly relates to the technical field related to resource scheduling, real-time indexes such as CPU utilization rate, memory occupancy rate and network bandwidth are acquired through a lightweight monitoring agent, and resource demands in the future 3-10 minutes are predicted by using an improved LSTM (including a cross-cycle attention mechanism), so that resource scheduling optimization is realized. The heterogeneous resource matching degree is calculated in combination with a graph attention network, a hierarchical scheduling strategy and a dynamic fault-tolerant mechanism are implemented, and the scheduling effect is evaluated through a multi-objective optimization function. The system comprises a distributed sensing terminal, a predictive analysis engine, a decision center and other modules, and supports federated learning, elastic capacity expansion and contraction and visual evaluation. The resource utilization rate can be improved, delay and energy consumption are reduced, and the method is suitable for heterogeneous resource scheduling scenes such as cloud computing and edge computing.

The invention discloses a power network attack chain dynamic deduction and intelligent response process method, system and device based on deep reinforcement learning and a medium, and belongs to the technical field of network security and power system protection. Multi-modal data is aligned and normalized, an event view cache is constructed, and the generalization detection capability on process camouflage and memory injection attacks is improved through a federated learning collaborative detection mechanism; based on the event view cache and historical threat intelligence, generating a dynamic attack knowledge graph, constructing a deep reinforcement learning model taking the attack knowledge graph as an environment, calculating an attack influence index by using a Bayesian network, and generating a differentiated security response instruction; and realizing attack path backtracking and attack source positioning based on the attack knowledge graph. According to the method, multi-modal data fusion analysis and strategy adaptive updating are realized, and the attack chain identification accuracy and evidence chain construction integrity are remarkably improved.

Network computing equipment may receive one or more radio parameter measurement reports generated by one or more user equipment. A report may comprise radio parameter values measured by user equipment corresponding to signals transmitted by a radio access network node and time information corresponding to the report. The computing equipment may perform temporal analysis based on time information corresponding to one or more reports and may generate a temporal graph based on the time information. Edges of the temporal graph may correspond to connection activity of user equipment that generated the reports with respect to one or more radio access network nodes. The edges may be ranked using a graph-based model and optimally combined with anomaly scores, determined based on the measured radio parameter values, into combined anomaly scores that may be compared to a criterion to determine that a radio access network node is a fake base station.

The invention relates to the technical field of intelligent logistics, and discloses a logistics supply chain optimization system and method based on artificial intelligence, and the method comprises the steps: constructing a heterogeneous vehicle characteristic vector, and building a digital twinborn model; calculating a matching relationship between the task and the vehicle based on a graph attention network; decomposing the scheduling problem into multi-level sub-problems by adopting a hierarchical reinforcement learning algorithm; constructing a decentralized collaborative decision-making system; a task exchange protocol based on game equilibrium is realized; according to the method, the problems of heterogeneous vehicle resource mismatching, centralized decision response lagging, low multi-level decision main body cooperation efficiency and insufficient large-scale task cooperation are solved, and the logistics distribution efficiency and the service quality are improved.

The invention discloses an intelligent anti-counterfeiting color box printing positioning control method, and relates to the technical field of printing positioning control, and the method comprises the steps: dividing a color box layout into a plurality of detection regions, setting a cross-shaped and circular combined positioning mark in each region, and collecting positioning information in real time through a high-resolution image sensor to establish a feature database; extracting regional deformation and position features by using a multilayer convolutional neural network, calculating actual deviation by combining a residual network, constructing a mapping model of process parameters and position deviation, and establishing a differential compensation model by using a deep reinforcement learning algorithm; and the compensation parameters are distributed to a servo control unit, a motion control instruction is generated, a compensation effect evaluation index is adopted to carry out real-time evaluation on position deviation, a decision model is established to realize compensation parameter adaptive optimization, and regional compensation adaptive adjustment is realized. According to the method, intelligent monitoring and optimization of the anti-counterfeiting printing process are achieved, and the printing quality stability and the production efficiency of the anti-counterfeiting color box in batch production are greatly improved.

The invention relates to the technical field of industrial mechanism models, in particular to a shock absorber performance optimization control method based on model fusion, which comprises the following steps: extracting a low-frequency disturbance state variable and inputting the variable-topology industrial mechanism model to generate a nominal reference state trajectory; utilizing a depth state observation network fused with energy passivity constraint to calculate a non-linear model mismatch compensation amount and an adaptive weighting parameter; performing dynamic fusion on the nominal reference state trajectory and the compensation amount based on the adaptive weighting parameter to generate a generalized state estimation value; and executing dynamic multi-objective optimization based on the generalized state estimation value, and generating mixed mode control input acting on an execution end. According to the invention, through adaptive fusion of a mechanism model and a data driving method, physical consistency, calculation real-time performance and robustness of a control process are considered.

The invention discloses a traffic accident detection method based on FFC and GCSA models. The method is innovatively improved based on a YOLOv8 network model. Firstly, a multi-scale feature fusion module FFC is designed in Backbone to replace an original C2f module, and the capability of fusing the multi-scale features of the network can be further improved on the premise of not increasing the network calculation amount, so that the network can fuse the multi-scale features on the level of finer granularity; then, a GCSA attention mechanism is introduced between a C2f module and a Deect module of a Neck layer, and the recognition capability of the model for key features in a complex environment is remarkably enhanced; and finally, establishing a loss function Focal-EIoU Loss for the improved YOLOv8 network structure, so that the network classification detection capability is improved, and the generalization capability of the model is improved. In a specific implementation process, a traffic accident image data set covering multiple scenes is constructed, and specialized data preprocessing is performed; then, end-to-end training and parameter optimization are carried out on the Traffic-YOLO network model obtained after improvement; and finally, integrating the optimized model to a traffic accident detection system for real-time target detection. Compared with the prior art, the method effectively improves the accuracy and robustness of traffic accident detection in a complex scene, and has important practical significance.