15 results about "Network segmentation" patented technology

The application relates to a GPU-based high-performance multi-party secure computation training method and system, and particularly relates to the field of multi-party secure computation protocols.The application aims to provide a multi-party secure computation training framework with higher parallelism, so as to realize parallelism between different layers of a neural network in a manner of combining data parallelism and model parallelism, and improve the data throughput speed of a training process.The method is a multi-party secure computation training system based on a pipeline flow training method, as shown in Figure 1, the method is designed according to the characteristics that the bottlenecks of linear computation network layers and nonlinear computation network layers in the MPC model training process are calculation and communication respectively, a pipeline flow training method is designed, parallelism between sub-networks is realized, and an optimal sub-network segmentation algorithm is realized to balance the training load between each sub-network.The application provides a multi-party secure computation training framework with higher parallelism, parallelism between different layers of a neural network is realized in a manner of combining data parallelism and model parallelism, and the data throughput speed of a training process is greatly improved.

The application relates to the technical field of intelligent traffic and infrastructure monitoring, and particularly discloses a highway infrastructure intelligent monitoring and management system based on wireless communication. The system comprises a wireless sensor node cluster, a relay coordination node and a network management server. The system combines a hybrid routing mechanism of distributed greedy forwarding and centralized optimal path calculation, and dynamically adjusts a routing strategy according to data service priorities, so that balanced distribution of network load and energy consumption is realized, low-delay and high-reliability transmission guarantee is provided for high-priority service data packets, the network life is prolonged, and the timeliness of key information reporting is improved. The system combines a hybrid routing mechanism of distributed local decision and centralized global optimization, and fundamentally solves the problems of energy consumption hotspots and network segmentation caused by static networking.

A network is partitioned into multiple domains and managed by a network controller. The network controller stores segmentation policies defined in terms of user groups, wherein members in a user group are identified by IP addresses or IP prefixes. An administrator controls which user groups communicate with each other and in which domains by associating segmentation policies with domains. Classification sources external to the network controller inform the network controller of changes to the membership of the user groups. The network controller uses the group membership information and segmentation policy / domain assignments to generate traffic policies. Each traffic policy is specific to a domain; the traffic policy is generated only from the segmentation policies of the domain, and consists of segmentation policy rules associated with the domain and members in the group's associated with those policy rules.

The application provides a deep neural network segmentation strategy optimization method and system in a multi-unmanned aerial vehicle scene, which is suitable for a distributed neural network training scene of user equipment, unmanned aerial vehicle nodes and base stations. The method divides the deep neural network into a front-end sub-network, a middle sub-network and a rear-end sub-network, which are respectively deployed on the user side, the unmanned aerial vehicle side and the base station side. A time slot parallel computing and transmission mechanism is constructed, so that different nodes perform forward propagation, backward propagation and hidden layer data and gradient transmission in parallel in the same time slot, thereby effectively reducing the overall training time delay. For the discrete layer number allocation problem of the middle sub-network among multiple unmanned aerial vehicles, a solution space continuous modeling method is introduced, the network segmentation strategy is converted into a continuous variable optimization problem, and a differential evolution algorithm is used for solving, so that the layer number of the sub-network of each unmanned aerial vehicle node is adaptively configured, and the parallel degree and resource utilization efficiency of the system are improved.

The application discloses a kind of high-efficiency double-path encoding network segmentation methods for lung nodule automatic segmentation, comprising the following steps: S1, collect lung nodule CT image, and the collected lung nodule CT image is pretreated;S2, the data set after processing is proportionally randomly divided into training set, verification set and test set, and the lung nodule CT image of training set and verification set is used to train the high-efficiency double-path encoding network segmentation model for lung nodule automatic segmentation constructed;S3, the lung nodule CT image on test set is introduced into the double-path encoding network model trained, and the segmentation result is obtained.The method combines the advantages of double-path encoding and global modeling, significantly reduces the model complexity while improving the overall segmentation performance, solves the problems of high model complexity, insufficient feature expression, inaccurate nodule boundary positioning and limited adaptability to small targets and multi-scale structures in existing lung nodule segmentation methods.

This disclosure describes techniques for enabling multiple subnets across multiple fabric sites and associated with multiple network segments (e.g., virtual networks (VNs)) to communicate with each other using a shared network infrastructure, such as a service provider network. In some cases, the techniques described herein include using a common transit VN (e.g., a common transit VN with or without a common firewall) in the shared network infrastructure as well as border devices that enable switching traffic between the common transit VNs and segment VNs (e.g., subscriber VNs) for data transmission to and / or from the common transit VN. In some cases, a border device maintains two types of mapping entries (e.g., map-caches): transit mapping entries and local mapping entries. A transit and a local mapping entry may be configured to represent (e.g., installed to program) forwarding information for packets received on a transit VN and on a segment VN, respectively.

The present application relates to the technical field of steep pulse shaping of pulse forming circuit, and particularly relates to a steep pulse energy shaping method based on a pulse forming circuit, which comprises: charging an energy storage capacitor group in a preset charging period, and initially gating a shaping network composed of PFN segments and Blumlein transmission line segments and an impedance gradient matching network; triggering a shaping switch to output a test pulse when a charging voltage reaches a target value, obtaining an incident component and a reflected component through an output sampling branch and generating a reflected characteristic quantity; the reflected characteristic quantity performs gating update on the matching network segment impedance configuration and the shaping network segment combination according to the minimum reflection criterion, and triggers a retest test pulse to obtain a retest reflected characteristic quantity; triggering a working pulse output when the retest reflected characteristic quantity is less than or equal to a reflected judgment threshold, and obtaining a steep pulse energy shaping result at a load end. The present application improves impedance matching consistency and waveform stability through reflected quantity closed-loop gating.

The invention provides an industrial process fault diagnosis method based on a Kolmogorov-Arnold network piecewise linear spline edge function, and belongs to the technical field of industrial artificial intelligence and process safety monitoring. The method comprises the following steps: performing data preprocessing on original industrial sensing data through standardization and a sliding window to form a fixed-length sample sequence, and inputting the fixed-length sample sequence into a KAN edge function layer; the method comprises the following steps: adding a linear transformation item and a first-order piecewise linear spline item to obtain a low-dimensional implicit representation; then, outputting a fault category through a lightweight classification network; sparse regularization is introduced in the model training process to promote part of edge function coefficients to converge to zero, after training is completed, the importance ranking of all variables can be obtained through edge function coefficient analysis, and interpretable analysis of a fault judgment basis is achieved in combination with a representative edge function response curve. The method is high in diagnosis precision, the model is light and efficient, the result is transparent and explainable, and variable-level and interval-level accurate contribution degree tracing can be achieved.

The application provides a kind of appearance detection method of embedded fingerprint module based on semantic segmentation, comprising: collecting fingerprint module sample image, positioning, labeling of glue overflow, bump;Defect semantic segmentation network model building;Establishing defect level quality evaluation module;And fingerprint module product defect online logic detection, collect actual fingerprint module product image to be detected, through global fixed threshold segmentation and local dynamic threshold positioning defect area to be detected product, call fingerprint module semantic segmentation detection model file, predict output glue overflow, bump defect, introduce defect level quality evaluation index to determine OK / NG product.The method comprehensively analyzes the characteristics of glue overflow and bump, builds a deep learning neural network segmentation network, extracts the occurrence area of glue overflow and bump with high precision, and finally applies the defect level quality evaluation module to comprehensively analyze the current product, improves the appearance quality safety stability of fingerprint module.

This invention belongs to the field of image processing technology, specifically relating to an image-guided circuit breaker assembly method and system. The method includes: processing video of the assembly process using a neural network segmentation technique to accurately separate the energy storage spring and the trip hook; quantifying instantaneous strain by calculating the component of the optical flow field along the principal direction of the spring during its dynamic compression process, thereby obtaining a spring strain index to evaluate the smoothness of the process; calculating the hook engagement confidence score to evaluate the final accuracy by detecting its key points and analyzing its topological geometric features such as engagement distance and angle after assembly; multiplying and fusing the spring strain index and the engagement confidence score to obtain a joint assembly quality score, and determining whether the assembly is qualified, thus achieving a comprehensive and automated evaluation of assembly quality.

This invention discloses a multi-target detection pre-tracking method based on TransUnet segmentation guidance. It constructs a motion model and uses maximum accumulation of values ​​(M-calculation) to non-coherently overlay multi-frame measurement data. A TransUnet network is used to extract global context features, generating candidate regions for potential target trajectories. Based on Huber loss, M-estimation is used to fit potential target trajectories, extracting the starting point, direction of motion, and velocity of the potential target trajectories in parallel. Prior information is used to construct a local state space under kinematic constraints. Trajectories of all candidate regions are precisely estimated using a dynamic programming pre-tracking algorithm within the local state space. Track backtracking and threshold detection are then performed. This invention uses DP-TBD to perform threshold decisions for all candidate targets within the local state space, eliminating false track interference from network segmentation and ensuring tracking performance for weak targets. Simultaneously, the use of deep learning priors effectively reduces the state search space, significantly lowering computational overhead.