50942results about "Internal combustion piston engines" patented technology

Provided is a road crack detection method, a medium and a product. The lightweight YOLO-MCS road crack detection model is constructed based on an improved yolov8 network. A method of improving the yolov8 network includes: replacing a feature extraction network backbone with a lightweight convolutional neural network MobileNetV3, and embedding a Coordinate Attention Mechanism (CA) module in the lightweight convolutional neural network MobileNetV3; adding a small target detection layer and a Squeeze and Excitation (SE) module at a neck end; and introducing a loss function of power Intersection over Union (IoU) into a head prediction structure. The road crack detection method overcomes the problem that the existing detection algorithm has low detection precision for small cracks and is difficult to be applied to edge devices with limited computing resources.

The invention provides a real scene three-dimensional modeling method and system fusing laser point cloud and images, and belongs to the technical field of urban three-dimensional modeling. According to the method, laser point cloud data and multi-view image data of a target scene are acquired through a synchronous acquisition device, and joint calibration is performed on a laser radar and a camera to generate space-time synchronization parameters. And a dynamic weight registration algorithm is utilized to dynamically adjust the registration weight according to the point cloud density distribution and the image texture complexity so as to realize the multi-level registration of the laser point cloud and the image data. The registered data is used for constructing a three-dimensional geometric model and generating a seamless texture map, meanwhile, semantic-driven hole repairing is carried out on a model missing area, and finally a complete three-dimensional model is output. According to the real-scene three-dimensional modeling method and system fusing the laser point cloud and the image, the limitation of single data source modeling is solved, and real-scene three-dimensional modeling with high precision, high integrity and real-time updating is realized.

The invention discloses a visible light and infrared image fusion method based on cross-modal dynamic collaboration. The method comprises the following steps: respectively extracting texture detail features of a visible light image and thermal radiation features of an infrared image through a visible light encoder and an infrared encoder; spatial alignment and channel complementarity optimization of cross-modal features are realized by using a heterogeneous attention collaboration module; and performing layered fusion on deep semantics and shallow detail features through a dynamic gating multi-scale decoder to generate a high-resolution fusion image. According to the method, the problems of feature dislocation, detail loss and unreasonable fusion weight distribution caused by modal difference in the prior art are solved, the detail fidelity, the thermal target saliency and the complex scene adaptability of the fusion image can be remarkably improved, and a high-robustness fusion result is provided for low-illumination environment perception and multi-modal target recognition.

The invention discloses a building facade defect intelligent detection method. Comprising the steps of multi-illumination condition adaptive acquisition and preprocessing, multi-view image geometric correction and registration, multi-scale invariant feature extraction and characterization, defect candidate region segmentation under a complex background, defect topological structure reconstruction based on physical constraints, defect parameter accurate quantization and classification, result visualization, report generation and the like. According to the invention, through the technologies of regional adaptive illumination compensation, view angle invariant feature transformation, class defect interference suppression, fracture connection based on physical constraints, sub-pixel-level width measurement and the like, the sub-pixel-level width is measured; the technical problems of defect feature enhancement, complex background interference suppression, multi-view defect unified characterization, defect topological structure reconstruction, accurate parameter quantification and the like under the multi-illumination condition are solved, and the accuracy and the automation degree of building facade defect detection are improved.

The present invention relates to the field of vehicle data processing, and in particular to an intelligent decision support method and system for vehicle-mounted sensor data fusion. The method comprises the following steps: obtaining vehicle multi-source operating status monitoring data based on vehicle-mounted sensors; performing vehicle status heterogeneous fusion and operating status multi-period change analysis on the vehicle multi-source operating status monitoring data, thereby obtaining vehicle operating status evolution trend characteristics; obtaining vehicle driving recorder monitoring images and vehicle dynamic environment monitoring parameters, and performing multi-target visual detection and dynamic operating scene fitting to construct a vehicle operating status scene graph; performing three-dimensional scene point cloud modeling on the vehicle operating status scene graph, and performing dynamic vehicle trend rendering according to the vehicle operating status evolution trend characteristics, thereby obtaining a dynamic vehicle scene twin model. The present invention realizes efficient and intelligent vehicle data decision analysis, and improves the safety and stability of vehicle driving.

The invention provides an intelligent traffic decision-making method based on multi-source sensor data, and the method comprises the steps: obtaining vehicle sensor data and road information, and obtaining a standardized traffic flow feature vector through data synchronization, fusion, feature extraction and dimension reduction processing; inputting the standardized traffic flow feature vector into a traffic flow prediction model, and predicting the traffic flow through a vehicle dynamics model constraint state transition matrix; according to the predicted traffic flow and the real-time road condition information, the road congestion state is judged, and a real-time congestion map is generated; in combination with the real-time congestion map, the road network speed limit rule and the starting and ending point position information, an optimal path is planned, and path smoothing processing is performed through a driving behavior mode; and constructing a vehicle intelligent decision-making system, integrating traffic flow prediction, congestion detection and a path planning model, and outputting a vehicle driving strategy according to real-time vehicle sensor data.

The invention discloses a trajectory optimization method based on a multi-source heterogeneous positioning data fusion algorithm, and relates to the technical field of intelligent navigation and high-precision positioning, multi-modal data are acquired through a multi-modal sensor array, positioning redundancy of scenes such as tunnels and indoor scenes is enhanced, a weight distribution strategy is dynamically adjusted through an Actor-Critic network architecture, and the positioning accuracy is improved. The state space input comprises an environment semantic tag, a historical error sequence and a real-time noise variance, the output action space is continuous weight distribution of each data source, a multi-target reward function optimization strategy is combined, scene adaptability is realized, a local SLAM map, inertial navigation error parameters and a weight distribution strategy are shared in real time based on a V2X protocol, and the real-time performance of the system is improved. According to the method, a single device accumulative error is compensated by using adjacent vehicle data, a terminal locally trains an error compensation model, parameters are uploaded to a cloud end through differential privacy encryption, the cloud end adopts a FedAvg algorithm to aggregate a global model and issue the global model, the error difference between devices is inhibited, and dynamic road network updating and scene differentiation model distribution are supported at the same time.

The invention relates to the technical field of path planning, and discloses a path planning and control method and system for a container port automatic driving vehicle. The method comprises the following steps: acquiring environment data through a multi-source sensor, and obtaining a harbor district digital twinborn model through space-time registration fusion; a three-layer progressive planner is used for optimizing and generating a main and standby path scheme based on the model; predicting a dynamic target trajectory by using a time sequence attention network to obtain situation assessment; processing the evaluation data through a distributed predictive control algorithm to generate a control instruction; and monitoring a motion state through adaptive security constraints to form a security policy, and continuously optimizing by using deep reinforcement learning. According to the method, safe and reliable operation of the automatic driving vehicle is realized in a complex and changeable harbor district environment, and particularly, accurate path planning and control decisions can be made in the face of dynamic obstacles.

The invention relates to a heterogeneous multi-unmanned aerial vehicle cooperative path planning method based on multi-agent deep reinforcement learning, and solves the problem that a traditional path planning algorithm is difficult to deal with a dynamic planning problem and an agent cooperation problem compared with the prior art. And the existing multi-unmanned aerial vehicle cooperative path planning algorithm has the defects of insufficient feature extraction capability, low experience learning efficiency and poor cooperative strategy flexibility in a dynamic uncertain environment. The method comprises the following steps: analyzing a multi-unmanned aerial vehicle cooperative path planning task; building an airspace reinforcement learning environment; introducing an unmanned aerial vehicle kinetic equation; modeling a multi-unmanned aerial vehicle decision model in multi-unmanned aerial vehicle cooperative path planning as a POMDP model; designing an MASAC-SEPR algorithm model to carry out path planning on multiple unmanned aerial vehicles, and generating a multi-agent collaborative path optimization network model; and training the multi-agent collaborative path optimization network model. According to the method, the efficiency and the accuracy of multi-unmanned-aerial-vehicle cooperative path planning are remarkably improved, and a reliable solution is provided for heterogeneous multi-unmanned-aerial-vehicle cooperative operation in a dynamic uncertain environment.

The invention discloses a method and a system for collecting decision data by an intelligent agent driven by a multi-modal large model. The method comprises the following steps: collecting original data of a predetermined sensor, calculating information entropy and a normalized mass score, and generating a fusion feature vector through self-adaptive multi-modal information fusion; time sequence sensitive multi-mode alignment processing is carried out on the image, and time sequence alignment features are obtained by combining timestamp embedding and a self-attention mechanism; based on this, decision distribution is generated by using a Bayesian neural network, uncertainty is evaluated, and a final decision and index set is generated; evaluating a task score in combination with task description, and making an execution plan according to a resource state; initializing the model, generating synthetic data, optimizing the model and storing experience. According to the invention, the efficient decision-making process of the intelligent trolley is realized, and the execution capability and task completion precision of the intelligent trolley are improved.

The invention discloses an obstacle avoidance method and device for unmanned aerial vehicle power inspection autonomous flight based on a laser radar and a storage medium, and relates to the technical field of unmanned aerial vehicle obstacle avoidance, and the method comprises the steps: configuring the laser radar and a visual sensor on an unmanned aerial vehicle, and collecting the three-dimensional point cloud data and image data of a flight environment in real time; optimizing the point cloud data to generate a three-dimensional environment model; and based on the three-dimensional environment model, classifying the obstacles into static obstacles and dynamic obstacles, if the obstacles are static obstacles, identifying the types of the obstacles through a CNN model and performing positioning, and if the obstacles are dynamic obstacles, clustering motion tracks through a K-Medoids clustering method. According to the invention, static obstacles can be identified and positioned, and motion trail prediction can be carried out on dynamic obstacles such as flying birds through a dynamic obstacle prediction technology, so that a preliminary flight path is generated, surrounding obstacles are sensed and avoided in real time in the flight process, and it is ensured that the unmanned aerial vehicle can be smoothly executed in an electric power inspection task.

SOLUTION: A semi-supervised medical image segmentation method based on a visual language model includes the steps of: obtaining a medical image; inputting an unlabeled image and a text description into a visual language model, and obtaining a text-guided mask based on obtained dense image embedding and text embedding; inputting a labeled image into a student model, and calculating supervised loss by using obtained labeled image prediction; respectively inputting the unlabeled image into the student model and a teacher model to obtain unlabeled image prediction and a pseudo label, merging the text-guided mask with the pseudo label, and calculating semi-supervised loss by using the merged pseudo label and unlabeled image prediction; and performing medical image segmentation by using a trained student model on the basis of the supervised loss and the semi-supervised loss.EFFECT: A target segmentation region can be accurately identified by using advantages of text descriptions.SELECTED DRAWING: Figure 1

The invention discloses an intelligent driving multi-sensor fusion data processing system, which belongs to the technical field of fusion data processing and comprises a data preprocessing unit, a data fusion unit and a data post-processing unit. The data preprocessing unit normalizes the original data of the laser radar, the camera, the millimeter wave radar and the ultrasonic sensor through a standardization, cleaning, noise reduction and calibration synchronization module; the feature level fusion module deeply digs the characteristics of each sensor, and fuses geometric, visual, motion and close-range obstacle features by using a deep neural network; the decision-making level fusion module generates a driving instruction through weighted voting and fuzzy logic reasoning through classification and situation evaluation; and the data post-processing unit combines the vehicle and road condition optimization instruction, evaluates the risk, and stores the data feedback optimization system. The system can improve the sensing precision and decision reliability, enhances the expansion adaptability of the system, and provides guarantee for safe and efficient operation of intelligent driving.

The invention discloses a double-mechanical-arm adaptive motor neural network optimization system, and relates to the technical field of mechanical arm motion control, and the system is characterized in that a feature fusion sensing module synchronizes multi-sensor data through an FPGA, combines technologies such as YOLOv5s, Poi ntNet + + and Doppler frequency shift, achieves precise extraction and fusion of obstacle 2D / 3D features, and generates a real-time environment state matrix; the kinematic chain prediction module quantifies the obstacle motion probability based on historical trajectory clustering and Bayesian prediction, constructs a collision risk field, and dynamically evaluates the collision risk in the future 2 seconds; the planning module searches a low-risk path by adopting an improved A algorithm, optimizes a joint movement track in combination with quadratic programming, embeds kinematics and dynamics constraints, and ensures the feasibility and safety of the path; in addition, a correction mechanism and an energy consumption optimization mechanism are introduced, motion parameters are dynamically adjusted according to the real-time state of the mechanical arm and environment interference, and the obstacle avoidance performance and the operation efficiency of the mechanical arm are further improved.

A DA-YOLOv11-based unmanned aerial vehicle aerial image target detection method belongs to the technical field of unmanned aerial vehicle aerial image target detection, and comprises the following steps: S1, obtaining and preprocessing a disclosed unmanned aerial vehicle aerial image data set; s2, carrying out improvement on the basis of a YOLOv11 network model, and constructing a DA-YOLOv11 network model; s3, training an unmanned aerial vehicle aerial image target detection network based on DA-YOLOv11; and S4, inputting the test set for testing and evaluation. An expansion residual module is introduced, and a C3K2DWR module is provided, so that the receptive field of the model for a small target is enhanced, and the features of the small target are better captured; a self-adaptive pyramid pooling module is provided in combination with an improved context anchor point attention mechanism to enhance the expression of small target features, and global information interaction is enhanced at the same time; traditional bilinear up-sampling is replaced by a dynamic up-sampling operator, so that the feature fusion capability is enhanced; a WIoU loss function is adopted to adjust the shape and position information of a target frame, and the sensitivity of the model to a small target is improved.

The invention relates to the technical field of unmanned vehicle control, and discloses a smart city environmental sanitation unmanned vehicle path planning and real-time monitoring method. The method comprises the following steps: firstly, acquiring a multi-source environment sensing data set such as laser radar point cloud data, a camera image sequence and real-time traffic flow information; local road network features are extracted based on laser radar point cloud data, a dynamic target motion prediction map is generated according to a camera image sequence, and real-time traffic flow information is processed to generate a regional traffic efficiency evolution curve. And inputting the data into a path optimization model to generate an initial path sequence, dividing cleaning task priorities, fusing related data and generating a final path planning scheme through a reinforcement learning algorithm. In addition, operation state data of the unmanned vehicle are collected in real time, and a path correction instruction set is generated through an anomaly detection model to update the strategy network. According to the method, the rationality and the operation efficiency of the path planning of the environmental sanitation unmanned vehicle can be improved, real-time monitoring is realized, and the operation safety and the management intelligence level are enhanced.

The invention discloses a space-time adaptive matching system based on multi-objective reinforcement learning, and the system comprises a data fusion cleaning module, an intelligent path planning module, a multi-objective scheduling optimization module, an adaptive task optimization module, and a reinforcement learning driving module. Self-adaptive scheduling and path planning of the unmanned aerial vehicle in a three-dimensional space are achieved, intelligent aircraft task management is provided, operation safety in a complex airspace is guaranteed, multiple requirements of the low-altitude industry for safety and efficiency are met, meanwhile, dependence on static rules and expert experience is reduced, and the unmanned aerial vehicle route planning method is suitable for popularization and application. The method can dynamically adjust the strategy according to real-time feedback and environmental change, has high adaptability, and meets the core technical requirements of future low-altitude equipment in operation service, supervision and whole industry chain safety standardization systems.

A small-size vehicle detection deep learning model based on feature fusion of multi-scale modules is provided, which solves the problem of small-size vehicle image detection. The model includes a Backbone network, a Neck layer and a Head network, wherein a C2f_DCNv3 module based on the combination of deformable convolution v3 (DCNv3) and a cross stage feature fusion (C2f) module and an SPPF_LSKA module based on the combination of a spatial pyramid pooling fast (SPPF) layer and a large separable kernel attention (LSKA) module are introduced into the Backbone network; a C2f_SCConv module based on the combination of spatial and channel reconstruction convolution (SCConv) and a C2f module is introduced into the Neck layer; and a multi-scale kernel detection (MSK_Detect) module is introduced into the Head network.

The invention discloses an MRI medical image correction method and system based on a convolutional neural network and a computer readable storage medium, and the method comprises the steps: obtaining a brain MRI medical image, and carrying out the preprocessing; extracting lesion features of the brain lacuna infarction from the preprocessed MRI medical image through a convolutional neural network, and positioning a lesion area of the brain lacuna infarction; performing detail enhancement on the lesion area of the cerebral lacuna infarction, segmenting the lesion area and optimizing a segmentation result; and carrying out artifact restoration on the segmented image, and generating a three-dimensional focus visualization model based on the restored image. According to the method, through processing in multiple aspects such as quality evaluation, lesion recognition, segmentation, artifact repair and three-dimensional visualization of the MRI image, the lesion area can be recognized more efficiently and accurately, the image quality is improved, and an efficient and accurate brain lacuna infarction diagnosis auxiliary tool is provided.

The invention belongs to the technical field of positioning and navigation, and discloses a high-reliability positioning and navigation method for an air-ground cooperative unmanned system. According to the method, a negative obstacle is accurately extracted by using an RANSAC algorithm and a region growing method, a two-dimensional grid map is generated through three-dimensional point cloud re-projection, a negative obstacle region is marked on the map, and a prior map is provided for an unmanned vehicle. The unmanned vehicle constructs a local map in combination with a sensor of the unmanned vehicle, and the local map is matched with a prior map through an ICP algorithm to accurately position a global pose. According to a priori map and real-time sensor data, a grid cost map containing static obstacles, dynamic obstacles and negative obstacles is generated, a TSCNN predicts dynamic obstacle tracks according to time sequence data, map cost values are updated, an A * algorithm carries out optimal path planning based on the updated dynamic cost map, and the optimal path planning is carried out based on the updated dynamic cost map. And dynamic obstacles and negative obstacles are avoided in real time. According to the invention, efficient, accurate and safe navigation and obstacle avoidance can be realized in a dynamic environment and a complex terrain.

The invention relates to the technical field of AGV path planning, in particular to an intelligent obstacle avoidance and path optimization method and system for an AGV. The method comprises the following steps: obtaining an omnibearing image of an AGV activity area; performing static obstacle recognition on the omnibearing image to construct each static obstacle model; performing scene three-dimensional structure perception on the omni-directional image, and performing three-dimensional mapping based on each static obstacle model so as to obtain a global perception scene model; obtaining a cargo transportation planning log; performing cargo global available path planning on the global perception scene model according to a cargo transportation planning log, and constructing a global available planning path; and carrying out dynamic path optimization based on the global available planning path, and carrying out dynamic obstacle potential collision point mining according to the dynamically optimized transportation path so as to generate a collision time point and a collision position point. According to the invention, through dynamic and efficient path obstacle avoidance planning, the accuracy and stability of AGV driving are improved.

The invention discloses a verification method and system of a chip simulation model, a medium and a program product, and relates to the technical field of chip verification, and the method comprises the steps: obtaining a simulation model and a performance parameter index of a to-be-verified chip; generating model training data based on the historical performance indexes, the corresponding historical use case set and the historical verification result, and performing model training based on the model training data to obtain a use case generation model; inputting the performance parameter indexes to a case generation model to obtain a test case set and a test verification expectation; executing a plurality of test cases in the test case set based on the simulation model, and recording an actual response result of the simulation model; based on the actual response result and the test verification expectation, generating a test log, and determining an abnormal mark in the test log; and marking the fault features corresponding to the exception, taking the test cases as newly added training data, and optimizing the case generation model based on the newly added training data. According to the invention, the chip verification efficiency can be improved.

The invention provides a logistics support dynamic scheduling and emergency response system based on intelligent data fusion, and relates to the technical field of logistics support. Comprising a data acquisition module, a data fusion and processing module, a scene prediction and simulation module, a privacy protection collaborative scheduling module, a game optimization scheduling module, a transparent resource tracking module, a self-adaptive network optimization module and a moral priority scheduling module, according to the logistics resource dynamic scheduling system, global efficient optimization of logistics resource dynamic scheduling and intelligent analysis and strategy optimization of real-time data are realized through advanced technologies such as a variable component optimization algorithm, fusion of quantum annealing and a genetic algorithm, deep reinforcement learning and combination of Transform and a genetic algorithm; and the rationality of resource allocation and the scheduling efficiency are greatly improved.

The invention relates to an unmanned aerial vehicle intelligent dynamic path guidance method and system based on deep reinforcement learning. The method comprises the steps of obtaining three-dimensional terrain point cloud and obstacle information of a flight area, constructing an environment topological structure diagram, extracting features through a three-layer diagram convolutional network, and generating high-dimensional environment feature representation. And a deep reinforcement learning state space is constructed, and a multi-target reward function is designed. And initializing the deep reinforcement learning network by using [-0.1, 0.1] uniformly distributed random parameters to generate an initial path scheme. Optimizing through a dynamic planning algorithm, adjusting a node sequence according to a path length and a task dynamic threshold value, and calculating a target function value. And judging whether the current path scheme reaches the optimal balance point, if not, adjusting the weight of the reward function by using a gradient descent method, and regenerating the path scheme until the path scheme is optimal. And finally, converting the scheme into a waypoint coordinate and speed instruction, performing real-time monitoring during flight, and triggering re-planning when an obstacle exceeds a safety threshold, so as to achieve the purposes of safe flight of the unmanned aerial vehicle in a complex environment and the like.

The invention provides an intelligent traffic vehicle behavior modeling method and system based on collaborative learning, and relates to the technical field of unmanned driving, and the method comprises the steps: obtaining traffic data in a target region; performing data calibration processing according to the traffic data to obtain a collaborative perception data set; a vehicle space-time trajectory function is generated according to the collaborative perception data set, and a conflict domain space-time distribution diagram is obtained; according to the vehicle space-time trajectory function and the conflict domain space-time distribution diagram, probability distribution characteristics of the vehicle driving intention and a multi-workshop competition cooperation relation matrix are extracted, and a multi-dimensional interaction characteristic vector is obtained; and behavior modeling is carried out according to the multi-dimensional interaction feature vectors, and a vehicle behavior model representing interaction behavior rules and decision logic between vehicles is obtained. According to the method, through multi-vehicle cooperative sensing data calibration and dynamic fusion of road geometric constraints, the problems of single-vehicle sensor shielding and cross-vehicle data heterogeneity are effectively solved, and the global consistency of vehicle position and motion state estimation is remarkably improved.

The invention relates to the technical field of building information models, augmented reality, synchronous localization and map construction, and provides a deep learning-based SLAM-BIM augmented reality cooperative localization method and system, and the method comprises the steps: introducing a Transform time sequence feature extractor and a geometric relation graph, evaluating a dynamic distribution weight through combining with the confidence, achieving the cross-modal closed-loop detection, and obtaining an SLAM-BIM augmented reality cooperative localization result. A lightweight semantic segmentation network and a feature fusion module are utilized, a dense map with consistent geometric semantics is constructed, a space-time error propagation equation is constructed, online calibration is realized by means of BIM scale prior, an incremental fusion algorithm is designed, a global pose is optimized in combination with AR interaction, and the system fuses SLAM visual trajectory features and BIM semantic geometric features through a deep learning technology. According to the method, the problems that traditional SLAM accumulative errors are large and the BIM fusion precision is low are solved, robust positioning and map construction in a complex scene are achieved, the cooperation precision and real-time performance of SLAM and BIM are improved, and the method is suitable for AR scenes such as building construction and operation and maintenance.

The invention discloses a vehicle multi-modal trajectory prediction method based on an improved attention network, and belongs to the technical field of intelligent vehicle trajectory prediction, and the method comprises the steps: collecting historical trajectory data of a target vehicle and surrounding vehicles as an input sequence; secondly, constructing a vehicle multi-modal trajectory prediction model which comprises a motion feature extraction module, a space-time interaction module, a space-time fusion module and a trajectory output module; the motion feature extraction module uses a multi-scale convolution attention network and a gating circulation unit for processing, the space-time interaction module uses a dynamic graph attention network for extracting vehicle interaction information, and the space-time fusion module splices and fuses target vehicle motion features and space-time interaction features to obtain space-time fusion features; the track output module inputs the fusion features into a gating circulation unit, decodes the fusion features and then inputs the fusion features into a mixed density network, and multi-mode output of vehicle tracks is achieved; and finally, a proper loss function is selected for training, so that the prediction precision and the convergence speed of the model are improved.

The invention relates to the technical field of fault detection control, and discloses an automobile fault detection method and system fusing sensor data. The method comprises the following steps: modeling a multi-system sensor network through an automobile system topology mapping algorithm to obtain a sensor topology incidence matrix; performing synchronous processing on the sensor topological incidence matrix according to a vehicle dynamics constraint equation to obtain a synchronous multi-sensor data set; inputting the synchronous multi-sensor data set into a fault propagation path sensing network for fusion to obtain a vehicle system fault feature vector; detecting the fault feature vector based on an automobile physical model constraint optimizer to obtain a fault state recognition result; and tracing and controlling a fault state identification result according to the fault propagation atlas model to obtain a system adjustment control signal. According to the invention, the source identification accuracy of the complex coupling fault and the pertinence of control and adjustment are improved.

The present disclosure relates to a weakly supervised semantic segmentation method and apparatus based on an attention mask. The method comprises: inputting a sample image into an attention encoder, so as to obtain a global class token feature, an image classification result and two semantic segmentation results; then inputting the sample image into the attention encoder again, so as to generate an attention matrix and to randomly generate a target mask matrix; compensating for the attention matrix by means of the target mask matrix, so as to obtain a compensated attention matrix; on the basis of the compensated attention matrix, generating each local class token feature; and distinguishing positive and negative characteristics of the local class token features. Model losses not only include losses in terms of image classification and image semantic segmentation, but also comprises losses caused by distinguishing positive and negative characteristics of local class token features and performing comparative learning together with global class token features, and semantic segmentation of a model is supervised by means of introducing a plurality of losses, thereby improving the accuracy of semantic segmentation.

The invention discloses a road traffic flow prediction method based on a space-time mixed attention network, and the method breaks through the limitation of a conventional time sequence model and a single deep learning architecture based on the systematic analysis of urban road traffic flow space-time heterogeneity, periodic non-stationarity and road network topological relevance, constructs the space-time mixed attention network, and achieves the prediction of road traffic flow. Spatial heterogeneous correlation of road network nodes is captured through a graph convolution network, dynamic time sequence evolution characteristics of traffic flow are modeled by adopting a hybrid architecture, a residual attention mechanism is introduced to realize layer-by-layer refining of multi-scale spatio-temporal characteristics, and the overall architecture of the method has remarkable advantages in the aspects of spatial topology modeling and time dynamic capture compared with a traditional model. Feature decoupling learning is carried out on multi-source heterogeneous data such as weather and events, adaptive integration of environment sensitive features is realized through a parameterized gating fusion strategy, and the prediction error fluctuation amplitude in an extreme weather scene is reduced by 34.8%.