14 results about "Streaming algorithm" patented technology

The application discloses a kind of based on dynamic receptive field and adaptive feature fusion plate defect detection deep learning method, to improve the detection precision and robustness of complex background and small scale defect under industrial environment.The proposed method is based on YOLOv10, dynamic receptive field module and adaptive feature fusion module are introduced in its architecture.Dynamic receptive field module uses local features and context information to dynamically adjust the size of convolution receptive field, to enhance the perception ability of different scale target;Adaptive feature fusion module realizes the weighted fusion of multi-layer features through multi-layer perception mechanism, improves feature expression ability and target recognition performance.The experimental results show that the method of the application is superior to existing mainstream algorithms in detection accuracy and real-time performance, has good engineering adaptability and industrial application prospect, and is suitable for furniture manufacturing, building material processing and other plate detection scenarios.

This application provides a method for configuring video stream algorithms, comprising: extracting visual features, spatiotemporal features, and semantic features of the input video stream; fusing these features to generate a first scene feature of the video stream; and generating a configuration algorithm for the video stream based on the first scene feature and a pre-constructed knowledge graph representing the relationship between scene features and algorithms. The technical solution of this application can effectively solve the problems of high configuration cost, poor scene adaptability, knowledge transfer gaps, and low configuration efficiency of video stream algorithms in smart city scenarios in existing technologies.

The application discloses a UAV robust path planning method and system for uncertain dynamic environment, relates to the field of path planning, and realizes real-time tracking of dynamic obstacles by using EKF, and constructs a probability repulsive field based on Mahalanobis distance, effectively solves the sensing vulnerability and static model limitation of a traditional APF algorithm, and provides a robust dynamic obstacle avoidance guide for random tree expansion; secondly, a fuzzy logic adaptive step (FLC-AS) module is designed; the module intelligently adjusts the expansion step according to the risk degree of a local environment and an exploration stage, and realizes dynamic balance of exploration safety and search efficiency; comparison and analysis of the method and various mainstream algorithms show that the PFLS-RRT* algorithm has excellent adaptability in a complex environment, and achieves the best comprehensive performance in key indexes such as path quality, planning efficiency and robustness.

The application discloses a multi-target path planning method of a power station inspection robot. Path planning is a key link for operation of the power station inspection robot system, and the performance of the path planning is closely related to equipment inspection efficiency and energy consumption ratio. Current mainstream algorithms generally have defects such as slow convergence speed, poor safety and low planning efficiency. Compared with the traditional method, the application significantly improves the path planning efficiency and path quality by simultaneously optimizing multiple targets, and enhances the stability and safety of the path.

The application relates to a Shenwei intelligent acceleration card system based on a Chiplet technology, a Chiplet architecture chip taking a Shenwei processor and a domestic FPGA as cores, being capable of supporting mainstream AI algorithm models and accelerating an inference process through cooperation of the powerful double-precision floating point and integer calculation power of the Shenwei processor, the powerful parallel computing capacity of the FPGA and the high-flexibility low-latency performance of the Chiplet architecture, and being applicable to domestic computers, servers, intelligent products and the like. The devices used in the application are all domestic, and have a higher self-controllable level; the Chiplet architecture designed with a PCIe interconnection bus as a core has the advantages of high transmission bandwidth, high calculation efficiency and easy realization of an inter-chip interconnection protocol; the Chiplet research and development idea based on the domestic general-purpose processor and the FPGA can be expanded according to application requirements to face different application fields.

This invention belongs to the field of image processing technology, specifically relating to an online monitoring method and system for the drying morphology of instant goat milk powder. The method includes: acquiring images of the fluidized bed interior to construct a spatiotemporal fluidized image tensor; obtaining fluidization confidence using eigenvalue decomposition of the local spatiotemporal structure tensor matrix to determine the effective fluidized region; performing morphological opening operations on the effective fluidized region using a sequence of structuring elements, statistically summing gray values ​​and constructing a morphological pattern spectrum to obtain the morphological pattern spectrum entropy; obtaining the pulsating velocity vector using an optical flow algorithm, thereby obtaining the average particle pseudo-temperature and coefficient of variation; and obtaining the fluidization steady-state potential energy index based on the morphological pattern spectrum entropy, average particle pseudo-temperature, and coefficient of variation, and generating control commands. This invention achieves monitoring of complex and abnormal conditions such as the initial stage of dead bed formation and channeling through the fluidization steady-state potential energy index.

The application discloses a satellite slant beam inter-beam interference avoidance method and system based on cost flow and dynamic graph coloring, and accurately models the interference range of slant beams in a multi-satellite satellite hop-beam system, considers the influence of the slant beams when modeling the anti-beam inter-beam interference hop-beam pattern design problem, divides the dynamic graph coloring problem into two sub-problems of satellite-wave position matching and hop-beam time slot allocation which are coupled with each other, pre-solves the satellite-wave position matching problem by using a minimum cost maximum flow algorithm, obtains a satellite-wave position matching initial solution which maximizes the sum of satellite elevation angles of all wave positions in the system under the constraint of the maximum number of simultaneously opened beams of the satellite, and on the basis, combines a tabu search algorithm and a graph coloring algorithm to jointly optimize the two sub-problems, so as to further improve the inter-beam interference avoidance performance of the same hop-beam time slot opened beam. The application can obtain a multi-satellite hop-beam scheme with minimum inter-beam interference, and the algorithm has low complexity and is convenient for actual system implementation.

This invention belongs to the field of sensor and machine learning technology. It proposes a simulation test system and method for the mechanical properties of rock joints based on the principle of equivalent body force field. The main components include: constructing a geomechanical similarity model based on the principle of equivalent body force field, including a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress, which generates equivalent body force intensity inside the rock mass physical model; collecting multidimensional sensing data when the base moves relative to the interface of the rock mass physical model; automatically identifying the rock mass joint extension trajectory based on the multidimensional sensing data and using an improved U-Net machine learning model; obtaining displacement field data through an improved particle image velocimetry combined with an optical flow algorithm; using a Bayesian inference inversion model to invert the mechanical parameters of the joint surface; and evaluating the uncertainty of the inversion results using the Markov chain Monte Carlo method.

A method for detecting small targets in the airspace with global-local feature co-enhancement is presented, belonging to the field of airspace small target detection technology. This method strengthens the collaborative representation capability of global and local features by constructing a Contextual Feature Aggregation Module (CFFM), a Tiny Target Region Focusing Module (TTFM), and a Dynamic Feature Aggregation Module (DFAM). The network embeds CFFM in the backbone to integrate channel and spatial context information, introduces an additional branch and adds TTFM to enhance local perception of small targets, and sets DFAM to dynamically fuse multiple receptive field features before the detection head. This invention is validated using a self-built ASODD dataset and the publicly available VisDrone2019-DET dataset. Compared with existing mainstream algorithms, the mAPsmall index for small target detection is significantly improved, solving the problems of low accuracy and high false positive / false negative rates in airspace small target detection. It is suitable for efficient detection of small targets such as drones, birds, and kites in complex airspace environments.

The present application relates to production scheduling and intelligent manufacturing technology field, and proposes a kind of assembly line scheduling optimization method considering generalized priority constraint, including constructing integer linear programming model, generalized priority relationship reinforcement and propagation and work station capacity improvement are carried out to the input instance of integer linear programming model;Calculate simple lower bound based on task processing time, lower bound based on priority graph longest path and lower bound based on maximum flow algorithm, take the maximum value of three as final lower bound;Define including action space, define state space to represent the change of current solution relative to global optimal solution and local optimal solution on target function value and allocation balance degree, design reward function to adaptively select neighborhood operation;Based on final lower bound and Q-learning dynamic selection module, build initial scheme, carry out batch movement iteration by neighborhood operation adaptively selected by Q-learning dynamic selection module;Based on the result of batch movement iteration local search, output assembly line scheduling scheme satisfying generalized priority constraint.

This invention discloses a cloud disk snapshot backup method based on IO pressure awareness and dynamic throttling. The method includes: real-time collection of disk IO performance indicators of production virtual machines and health status indicators of the underlying storage system; comparison of the collected indicators with preset thresholds to dynamically determine the current system IO pressure status; calculation and output of a backup task IO rate limit value in real time using a dynamic throttling algorithm based on the IO pressure status; and rate limiting of the backup data reading process based on the IO rate limit value to achieve dynamic throttling. This application combines business layer indicators with infrastructure layer indicators to construct a more comprehensive and accurate system pressure assessment model, rather than relying solely on a single dimension. Furthermore, without altering the existing hardware infrastructure, it achieves IO resource coordination and automatic arbitration between backup tasks and production operations through software intelligence.

The application discloses a kind of self-adapting low-light image target detection method, belong to computer vision field.The method is improved to YOLO11 model, first convolution layer of backbone network is replaced by adaptive gamma enhancement module AGE, SPPF module is replaced by scale intra-feature interaction module AIFI based on attention, and adopts lightweight efficient detection head LEHead to detection head.AGE module predicts gamma coefficient through neural network, based on the gamma coefficient, input image is adaptively enhanced in brightness, and its perceptibility is improved;LEHead detection head adopts double 3x3 grouping convolution to realize feature enhancement, while ensuring detection accuracy, significantly reduces model calculation overhead, LEHead detection head includes decoupled classification branch and regression branch, classification branch adopts point-by-point convolution and Sigmoid activation to output class probability, and regression branch adopts point-by-point convolution and distributed focal loss supervision to decode output target bounding box.Experiments prove that the application is better than existing mainstream algorithm on ExDark and DarkFace dataset, with high precision and high efficiency, strong generalization, and can be widely applied to automatic driving, intelligent security, unmanned aerial vehicle inspection and other scenes.

This invention discloses a multi-objective train scheduling method and system integrating Q-learning, aiming to solve the problems of low efficiency and insufficient multi-objective coordination in existing technologies for large-scale solutions. Its core is to transform train scheduling into a work-shop scheduling problem with congestion constraints, establishing a mixed-integer programming model with the objectives of minimizing delay penalties, passenger dissatisfaction, and the total number of operating vehicles. A two-layer vector encoding of scheduling ordering and vehicle allocation is adopted, combined with multi-block left insertion heuristic decoding adaptation constraints. A multi-objective evolutionary framework is constructed by improving the NSGA-II algorithm, incorporating integrated population selection and customized crossover operators. Q-learning is introduced to drive variable neighborhood search, and adaptive selection of neighborhood operators enhances search intelligence. Experimental verification shows that this method outperforms mainstream algorithms in both solution efficiency and solution quality in large-scale examples and real subway data. It can efficiently balance multiple objectives, adapt to dynamic train formations and passenger flow fluctuations, and has engineering application value.

The application relates to a cooperative unmanned aerial vehicle cluster operation path planning method and relates to the unmanned aerial vehicle planning field. The application acquires starting positions and target positions of unmanned aerial vehicle clusters, constructs a minimum cuboid space covering the starting positions and the target positions, grids the minimum cuboid space into discrete space by using a cube; a static space cost topology graph representing reachability and movement cost between grid units in the discrete space is constructed, the static space cost topology graph contains nodes corresponding to the grid units and edges between the nodes, the edges are endowed with weights representing the movement cost; the static space cost topology graph is converted into a directed resource graph, the directed resource graph is configured with directed edges describing feasible movement processes of the unmanned aerial vehicles, the directed edges provide movement resources with a flow of 1 for the unmanned aerial vehicles; on the basis of the directed resource graph, a network maximum flow algorithm is used to find a collision-free path; the collision-free path is screened and post-processed considering the movement cost to obtain an executable path of the unmanned aerial vehicle.