29 results about "Class discrimination" patented technology

The invention discloses a double-channel confrontation and progressive self-training driven cross-working-condition open set domain adaptive fault diagnosis method. In order to suppress an interactive negative migration effect, a three-stage two-way adversarial progressive self-training network DAPN is designed, separation of unknown target samples and feature aggregation of known samples are realized through cooperation of a gradient inversion layer and a two-way adversarial discriminator, and inter-class discrimination and intra-class aggregation are obviously enhanced. When the performance of the DAPN is evaluated in two open set domain adaptive scenes, for open set proportions of different intensity domain offsets and changes, the method always keeps high classification accuracy and unknown class detection rate, and the performance is obviously superior to that of a comparison baseline. According to the method, two-way confrontation and a progressive self-training mechanism are combined, so that accurate classification of known faults and effective identification of unknown faults are realized.

The invention discloses a rotating machine intelligent diagnosis method based on structured pruning and knowledge fusion distillation, and the method comprises the steps: training a teacher network through a training set, carrying out the structured pruning of a percentile threshold value on the teacher network based on the L2 norm calculation and normalization of a convolution filter, and generating a student network with a consistent structure. A KFD strategy including feature-level distillation and logit-level distillation is utilized to carry out deep supervision on a student network, and two types of distillation losses are weighted and fused, so that a student model still keeps relatively strong feature characterization capability and category discrimination capability under a high pruning rate. Asymmetric integer quantization is adopted for the trained student network, so that the reasoning overhead is reduced, and the embedded adaptability is improved. A general neural network operator IP core is arranged on an FPGA, efficient deployment of a quantitative student model is achieved, and low-power-consumption and low-delay real-time fault diagnosis is achieved. The method has the advantages of being high in precision, light in model weight, easy to deploy and the like, and is suitable for on-line monitoring of industrial field rotating machinery.

The application discloses a feature extraction method and system based on difficult sample mining and multi-granularity division, which is used for a cross-view geographical image retrieval task. The method first preprocesses cross-view street view images and satellite images, and uses a generative model to generate cross-view images, reducing the visual difference between different view images. Then a two-stage difficult sample mining model is constructed, including a sampling strategy based on geographical location and visual similarity, mining difficult negative samples in different ranges, and enhancing the inter-class discrimination ability. Then a multi-granularity feature division module is introduced, the image features are extracted through a ResNet50 backbone network, and the features are divided and fused according to different granularities to obtain rich and robust view-invariant feature representation. Finally, the satellite image to be retrieved is input into the trained feature extraction model, the features are extracted, and similarity matching is performed with the street view image library to obtain the cross-view retrieval result.

An image semantic segmentation method based on an FAESeg network belongs to the technical field of image semantic segmentation, a network structure used in the image semantic segmentation method comprises a feature map extraction network, a semantic feature map adjustment network and a category discrimination enhancement network, and a large kernel decoupling module LDM is introduced into the feature map extraction network and the semantic feature map adjustment network. The problem of inaccurate processing of boundaries of different semantic regions is solved; a category discrimination enhancement module CDEM is introduced into the category discrimination enhancement network, so that the category discrimination capability of foreground images is enhanced, and more correct semantic classification is realized. According to the method, the FAESeg network is used for carrying out semantic segmentation on the image, so that a relatively accurate segmentation result can be realized while relatively low computing power overhead is used, and advanced performance is obtained.

The present application discloses a remote sensing image feature quality evaluation method, including: determining remote sensing image feature quality evaluation indicators, the evaluation indicators including: inter-class discrimination and intra-class aggregation; determining the value range of inter-class discrimination and the value range of intra-class aggregation through calculation and feature map visualization to evaluate the evaluation indicators; establishing a quantitative evaluation strategy for remote sensing image feature quality for evaluating remote sensing image feature quality. According to the determined evaluation indicators, based on statistical analysis technology, the value range of inter-class discrimination and intra-class aggregation is determined, and a quantitative evaluation method for feature quality is established to evaluate the quality of feature extraction results. It achieves the extraction of features with both high inter-class discrimination and high intra-class aggregation. The goal of further improving feature quality can be achieved based on quality evaluation in the future, which has far-reaching significance for remote sensing image feature quality evaluation and remote sensing image information extraction.

The application belongs to the technical field of open set identification, and discloses an open set identification method based on a hierarchical visual semantic attribute space. The application can not only accurately identify whether a target belongs to a seen class or an unseen class, but also can explicitly determine the coarse-grained class to which the target belongs and the specific attributes possessed by the target. Compared with the traditional open set identification technology, the open set identification method based on the hierarchical visual semantic attribute space has significant advantages, greatly improves the explainability of the model, and enables the model to not only give a simple unknown class discrimination when facing unknown targets, but also provide more valuable information, thereby providing solid technical support for the effective deployment of the model in complex real-world scenarios.

The invention discloses a high-resolution remote sensing building extraction method. The method comprises the following steps: acquiring multi-scale feature data and shallow enhancement feature data; obtaining deep enhancement feature data; obtaining fusion features; and generating a building extraction result. According to the method, a shallow feature enhancement module is constructed in a shallow spatial information representation stage, so that a network can more accurately highlight building related channels in a complex scene, redundant background information is inhibited, and the category discrimination capability and the positioning reliability are improved; according to the method, a deep feature enhancement module is constructed in a deep semantic information representation stage, the continuity and semantic integrity of a building boundary are improved, the perception and extraction capability of a small-scale building is enhanced, and false detection and missing detection are reduced; according to the method, a lightweight edge feature reconstruction decoder is designed in a decoding stage, so that boundary blur and artifacts caused by traditional interpolation up-sampling are effectively suppressed, and decoding stability and boundary recovery quality are improved on the premise of reducing parameter quantity and calculation overhead.

The present invention provides a method for enhancing and recognizing small-sample plankton images and a method for building its model, belonging to the technical field of underwater image enhancement. First, plankton microscopic image data is acquired and sorted out to construct a standardized data set. Then, a small-sample plankton image enhancement and recognition model based on feature map reconstruction technology is designed. A deep convolutional neural network is used to extract features from the image, generate intermediate layer feature maps, and combine the feature map reconstruction mechanism to achieve class discrimination. The model integrates multiple key designs for cross-domain adaptation and structure perception, alleviates the distribution differences between different data domains, and also significantly improves the model's ability to represent the complex morphological structures of plankton. Finally, the model is trained and optimized end-to-end under the meta-learning framework to obtain the best model. Experimental results show that this method exhibits excellent classification performance on multiple plankton data sets and can achieve high-precision recognition under the condition of limited sample quantity.

The invention discloses a human body action recognition method based on decoupling and structured modeling of CSI (Channel State Information). The method comprises the following steps: 1, collecting CSI action sample data; 2, preprocessing the collected CSI data; 3, completely integrated empirical mode decomposition, adaptive noise and Hilbert transform are introduced to carry out action-environment component dual decoupling on the preprocessed CSI data; 4, constructing a human body action recognition model based on the capsule network; and 5, generating unified action characterization insensitive to position change from the decoupled action signals by using capsule network vectorization coding and a dynamic routing mechanism, and constructing an enhancement path and a reconstruction path to respectively improve inter-class discrimination and structural consistency. According to the method, position-independent human body action recognition is realized based on modal decomposition and structured modeling technologies, action recognition under different position conditions can be effectively adapted, and high accuracy is achieved during new position testing.

The invention discloses a family dangerous object identification method and system based on improved YOLOv8, and belongs to the technical field of computer vision and edge computing. Aiming at the problems of long tail distribution of dangerous objects in a family scene, missing detection of small targets, false alarm of similar objects and limited computing power of edge equipment, three improvements are carried out on a YOLOv8s model: a class prototype adaptive mechanism is introduced, and the long tail class discrimination capability is enhanced; a multi-scale feature enhancement module is added, and the small target feature extraction effect is improved; a coordinate attention mechanism is embedded, and the target positioning precision is enhanced. Furthermore, the detection result is subjected to semantic review by using a lightweight multi-modal large model, so that false alarms are inhibited. And efficient deployment of edge equipment is realized through a model quantification technology. Experiments show that the mAP50: 95 is improved by 2.84% on a self-built data set, and the recognition accuracy and the system reliability are remarkably improved while low delay is kept.

The invention discloses a lightweight vehicle detection method based on improved YOLOv10, and the method focuses on improving the precision and efficiency of vehicle target detection and reducing the consumption of computing resources. According to the method, by fusing a multi-scale local channel attention mechanism of a C2fMLCA module, the perception capability of vehicle feature details is enhanced; an improved SENetV2 module is introduced to optimize a PSA module in the baseline model, a PSASENetV2 module is constructed, channel and space attention are combined, feature expression is optimized, and the recognition accuracy is improved; dynamic sampling is realized by adopting a DySample module, a sampling strategy is adaptively adjusted, the calculation amount is reduced, and the reasoning process is accelerated. In addition, the optimized v10DetectLSCD detection head is combined with locality sensitive hashing and channel attention, so that the target positioning and category discrimination efficiency is improved. According to the method, the parameter quantity and the model calculation cost are reduced while the high detection precision is kept, and the method is suitable for vehicle identification under the condition that equipment is limited and has wide application prospects and practical value.

This invention relates to the fields of network security and deep learning, specifically a Tor website fingerprinting method for satellite internet. It achieves this by constructing a self-supervised contrastive learning framework to generate multi-view augmented samples; extracting temporal-granularity feature matrices; generating synthetic negative samples based on linear interpolation, which are then compared with positive samples to form contrast pairs; extracting trajectory representations; and jointly optimizing the results using cross-entropy loss and mask reconstruction loss. Compared to traditional website fingerprinting methods, this invention enhances the model's robustness to satellite-borne inherent noise caused by the Doppler effect. The weakly and strongly augmented views of this invention provide multi-scale semantic features, while the synthetic negative samples improve inter-class discrimination and reduce the false positive rate. The joint loss optimization framework allows the model to consider both global statistical characteristics and local burst features, significantly improving the accuracy and adaptability of website fingerprinting under the conditions of highly dynamic topology and limited annotation in satellite internet constellations.

The application discloses an interpretable image classification method with double-role decoupling and probability conservation redistribution, and aims at the problems that the existing CAPE type interpretable method excessively pays attention to local discrimination areas, the complementary areas are easily interfered by backgrounds, the explanation heat map lacks probability consistency and cannot explicitly distinguish the roles of areas, and the application constructs a discrimination branch and a complementary branch to respectively learn a class discrimination area and a target complementary area, realizes explicit spatial division of the double branches through a spatial mask router, adopts adaptive weight fusion of double branch results, and optimizes the spatial distribution of the explanation heat map under the premise of keeping the total probability of the class unchanged through a probability conservation redistribution mechanism. The explanation heat map generated by the application has discrimination, target integrity and probability consistency, effectively suppresses background interference, and is suitable for visual intelligent analysis scenes such as bird fine-grained identification and medical image auxiliary diagnosis which need interpretability.

The invention relates to the technical field of cerebral infarction risk prediction, and discloses a PCIS early risk prediction system driven by multi-modal data, and the system comprises a data obtaining module which obtains PCIS-related multi-modal data; the feature extraction module is used for extracting features of the images and the multi-omics data; the feature reconstruction module reconstructs missing features; the adversarial generation module is used for generating pseudo-omics features and pseudo-image features for eliminating heterogeneity among different modes through an adversarial learning network; the intra-modal comparison module is used for enhancing the category distinguishing capability of the pseudo-omics features and the pseudo-image features by using comparison learning; the cross-modal alignment module is used for aligning semantic consistency between cross-modal features through cross-modal comparison learning; the feature fusion module is used for fusing features by using an attention mechanism; and the prediction module is used for generating a PCIS early risk prediction result according to the fused features. According to the method, the reconstruction of the multi-modal missing features and the effective extraction of the interaction relationship between the features can be realized, and the accuracy and robustness of PCIS early risk prediction are improved.