Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

609results about How to "Improve inference speed" patented technology

Large language model reasoning optimization method based on cascade and speculative decoding strategy

The invention relates to the technical field of language model reasoning, and discloses a large language model reasoning optimization method based on cascading and speculation decoding strategies, which comprises the following steps: S1, inputting cue words of a user to a small language model family, and quickly generating a plurality of draft text tokens by using a cascading strategy; s2, inputting a current text sequence into the large oracle model, verifying by using a speculative decoding strategy, cutting off a rejected text token, and generating a text output token of the large language model; s3, the step S1 and the step S2 are cycled until the large oracle model generates the result representing the end of the generation task; gt, endgt; and a label. According to the method, a cascading strategy and a speculation decoding strategy are introduced, the advantages of the two strategies are combined, and a new speculation cascading technology is designed, so that the reasoning speed and efficiency are improved. The cascading strategy is formed by combining a plurality of language models with different levels and different parameter sizes, and the speculative decoding strategy is used for reducing waiting time by deducing in advance and generating part of results.
Owner:PIO CLOUD COMPUTING (SHANGHAI) CO LTD

Large language model reasoning acceleration method, device and system and medium

The invention discloses a big language model reasoning acceleration method, device and system and a medium, and the method comprises the steps: carrying out the offline quantification processing of a big language model, and obtaining a quantification model; speculating a plurality of candidate lexical elements in parallel in each decoding step by using the quantitative model according to an input sequence; inputting the plurality of candidate lexical elements into the large language model for verification; and taking the verified candidate lexical elements as the output of the corresponding decoding step until a complete output sequence is generated. According to the method, offline quantification processing is carried out on a large language model, parallel candidate lexical element speculation of the quantification model on an input sequence can be obtained only through few hardware resources, the parallelism of each decoding step is increased to achieve reasoning acceleration, the illusion problem generated by different models can be relieved while the reasoning speed is increased, and the reasoning efficiency is improved. Model precision is ensured, and high-precision and low-hardware-cost reasoning acceleration is realized.
Owner:CHINA PING AN PROPERTY INSURANCE CO LTD

Crowd counting method based on adaptive global perception and multi-scale feature fusion

The invention belongs to the technical field of computer vision, and discloses a crowd counting method based on adaptive global perception and multi-scale feature fusion, and a crowd counting model built in the method comprises an adaptive global perception module AGP, a multi-scale Mama aggregation module and a channel perception Mama decoder CMD. The AGP realizes efficient extraction of global information of the image, and improves the reasoning speed; the multi-scale Mama aggregation module effectively fuses the features of different scales, enhances the capturing ability of detail and global features, and improves the accuracy of crowd counting; the CMD enhances the relevance of information between channels, and further improves the counting precision of the model; the composite loss function ensures rapid convergence and high-precision performance of the network in a crowd counting task. According to the method, the model reasoning speed and the counting accuracy are ensured, meanwhile, the calculation resource consumption is remarkably reduced, and the light weight of the model is achieved.
Owner:SHANDONG UNIV OF SCI & TECH

Dynamic environment dense point cloud SLAM method and system based on YOLOv11 and ORB-SLAM3

The invention discloses a dynamic environment dense point cloud SLAM (Simultaneous Localization and Mapping) method and system based on YOLOv11 and ORB-SLAM 3. According to the method, real-time target detection and image segmentation technologies of a YOLOv11 model are integrated into an ORB-SLAM 3 framework, so that high-precision and robust visual SLAM in a dynamic environment is realized. A traditional convolution layer is replaced by using a balanced convolution method GSConv layer in a YOLOv11 model, and a traditional C2f layer is replaced by using a new feature fusion module VoVGSCS layer, so that a Neck structure of the YOLOv11 is improved, and a lightweight network model is realized. Experimental data prove that the pose estimation precision of the method in a dynamic environment is remarkably superior to that of an existing visual SLAM algorithm.
Owner:ZHEJIANG UNIV

Method for improving reasoning solution and training speed of deep operator network model

The invention belongs to the technical field of artificial intelligence and high-performance computing, and particularly relates to a method for improving the reasoning solution and training speed of a deep operator network model. According to the method, multi-level optimization is carried out on the PI-DeepONet model, and the reasoning solving speed and the training speed of the PI-DeepONet model are improved. According to the method, the SGEMM operator is customized, and compared with the best single-precision matrix multiplication in a basic linear algebraic library of a CUDA (Compute Unified Device Architecture) computing platform, the SGEMM operator can obtain a speed-up ratio of 1.5 times at most in a common model size range of PI-DeepONet. A fusion operator is designed according to the special structure of the modified MLP used in the model, the speed-up ratio of the end-to-end reasoning speed of the model is more than two times, and the calculation time of a graphics processor can be shortened by 5.7 times at most. According to the method, data parallel is used for carrying out distributed training on the model, a communication concealment strategy is used, and the expandability is kept at 83%-85%.
Owner:CHINA UNIV OF PETROLEUM (EAST CHINA)

Large language model reasoning acceleration method and device, equipment and medium

The invention relates to the field of artificial intelligence, and discloses a reasoning acceleration method and device for a large language model, equipment and a medium, and the method comprises the steps: obtaining a to-be-processed text in a reasoning request in response to the reasoning request input by a user; performing video memory demand prediction according to the to-be-processed text to obtain a video memory prediction result; distributing a corresponding initial video memory space for the large language model according to the video memory prediction result so as to execute a corresponding reasoning task on the to-be-processed text; when the large language model executes the reasoning task, monitoring an occupation index of the initial video memory space; and when the occupation index meets a preset condition, triggering a dynamic allocation mechanism, and allocating a new video memory space for the large language model according to the dynamic allocation mechanism. The corresponding initial video memory space is allocated by predicting the video memory demand in advance, the use condition of the initial video memory space is monitored in real time, and a dynamic allocation mechanism is triggered when resources are insufficient, so that flexible and controllable dynamic video memory allocation is realized, the waste of video memory resources is reduced, and the overall reasoning speed of the model is improved.
Owner:CHINA PING AN PROPERTY INSURANCE CO LTD

Super-resolution assisted remote sensing image small target detection method

InactiveCN119048912AAlleviate the problem of difficulty paying attention to small goalsEasy to detectGeometric image transformationCharacter and pattern recognitionImage resolutionHigh resolution image
The invention discloses a super-resolution-assisted remote sensing image small target detection method. The main body of the method is composed of a small target detection network, a super-resolution-assisted branch network and a loss function. The small target detection network comprises five sub-modules, namely a backbone network module, a channel attention weighting module, a self-attention coding module, a neck network module and a small target detection head. The backbone network module is used for extracting small target features in the remote sensing image; the channel attention weighting module is used for giving different weights to different channels of the image so as to enhance information of useful channels of the remote sensing image; the self-attention coding module is used for increasing global context information of small targets in the trunk features so as to improve the feature expression ability of the small targets; the neck network module is used for fusing backbone networks with different resolutions and enhancing semantic information and detail information of backbone features; and the small target detection head is used for outputting a prediction frame and prediction information on each feature layer. The super-resolution auxiliary branch network comprises a super-resolution reconstruction module and a parallel sensing module. The super-resolution reconstruction module is used for reconstructing the low-resolution features extracted by the backbone network into high-resolution features; and the parallel sensing module extracts detail features in the high-resolution image for subsequent calculation of super-resolution sensing loss. The loss function is composed of two parts of super-resolution perception loss and detection loss. The method can effectively extract the feature information of the small target in the remote sensing image, and improves the detection performance of the small target in the remote sensing image.
Owner:NANJING UNIV OF AERONAUTICS & ASTRONAUTICS

Industrial anomaly detection method and device based on multi-mode joint memory pool

The invention provides an industrial anomaly detection method and device based on a multi-modal joint memory pool, and belongs to the technical field of industrial anomaly detection, and the method comprises the steps: carrying out the multi-modal feature extraction of input RGB image data and point cloud data through a pre-training network; feature mapping and alignment operations are introduced to align point cloud features with RGB image features in space; two groups of intra-modal codec and two groups of cross-modal codec are introduced to perform intra-modal and cross-modal compression and reconstruction on the obtained RGB image features and point cloud features; training four groups of codecs by adopting a cosine similarity function; constructing a multi-modal feature memory pool; and adopting a nearest neighbor algorithm to calculate an abnormal score. According to the method, the RGB image and point cloud input are comprehensively utilized, abnormal areas of different scales and different forms can be accurately detected, and the universality and robustness are high.
Owner:TIANJIN UNIV

Population mobility OD matrix prediction method

The invention relates to a population mobility OD matrix prediction method, and the method comprises the steps: constructing a plurality of space-time diagrams with different adjacent matrixes according to historical OD flow information and urban region information; performing feature extraction on each time-space diagram by adopting a three-dimensional diagram convolutional network, and performing feature fusion on feature extraction results to obtain time-space extraction results; block embedding is carried out on the time-space extraction result, position coding and time coding are carried out on the result after block embedding processing, and input data are obtained through fusion; a Transform module is adopted to extract long-term time features from input data, and a decoder outputs a final prediction result. Compared with the prior art, the method has the advantage of high prediction accuracy.
Owner:TONGJI UNIV

Oral cancer multi-task diagnosis system based on Raman spectrum

The invention discloses an oral cancer multi-task diagnosis system based on a Raman spectrum, and belongs to the technical field of biomedical detection. Comprising an acquisition module and an oral cancer multi-task diagnosis module, and the acquisition module is used for acquiring Raman spectrum data; the oral cancer multi-task diagnosis module is used for processing the Raman spectrum data through a trained oral cancer multi-task diagnosis model so as to generate and output a tumor staging result, a lymph node staging result and a histological staging result; the oral cancer multi-task diagnosis model comprises a bidirectional Mama branch and a residual attention fusion branch, local feature information in the Raman spectrum data is extracted by using the residual attention fusion branch, and multi-scale features are generated by fusing global feature information extracted by the bidirectional Mama branch. Feature information in Raman spectrum data can be effectively extracted, the accuracy of oral cancer diagnosis is improved, and multi-task diagnosis of oral cancer is realized; the problems that multi-task common diagnosis cannot be realized and the diagnosis accuracy needs to be improved in the prior art are solved.
Owner:BEIJING INFORMATION SCI & TECH UNIV

Occupied grid map construction method, training method, device and electronic equipment

The invention relates to an occupied grid map construction method and device, a training method and device and electronic equipment. The occupied grid map construction method comprises the steps of obtaining multi-modal data of a target area; inputting the multi-modal data into an occupied raster map construction model, and extracting multi-modal data features based on the multi-modal feature data in an encoder of the occupied raster map construction model; in a decoder of the grid map construction model, obtaining fusion features based on the first object query and the multi-modal data features, predicting the occupancy probability of the grid based on the fusion features, and performing finer-grained grid reconstruction on the target grid based on the occupancy probability to obtain a second object query; and obtaining an occupied raster map of the target area based on fusion features obtained by the second object query and the multi-modal data features. According to the method, the occupied grid map is constructed based on the multi-modal data, the calculation amount can be reduced through the construction mode, and the requirements for the calculation power and the storage amount are low.
Owner:CHONGQING CHANGAN AUTOMOBILE CO LTD

Vehicle driving track prediction method based on large language model

The invention discloses a vehicle driving track prediction method based on a large language model, and belongs to the technical field of automatic driving. The invention aims to provide a vehicle driving track prediction method based on a large language model, which is used for inputting a vehicle track into the large language model in a text format and evaluating the capability of the large language model in understanding the time dynamic aspect of a traffic agent. According to the method, tracks of different vehicles in a traffic scene are encoded by using a track encoder, track encoding features are mapped into a text embedding space of a large language model by using a reprogramming adapter, and the encoding features are translated into lexical combination which can be understood by the large language model. According to the method, powerful support is provided for realizing reliable and safe driving decisions by the large language model.
Owner:JILIN UNIVERSITY

Vehicle-mounted large model deployment method based on mixed granularity knowledge distillation

The invention discloses a vehicle-mounted large model deployment method based on mixed granularity knowledge distillation. The vehicle-mounted large model deployment method comprises the following steps: performing mask initialization on coarse particle and fine particle structures in a pre-trained large model; initializing low-rank matrixes, wherein the matrixes are used for replacing original weight parameters of the model for updating in the distillation training process; setting a model loss function including knowledge distillation loss and regularization loss; in the model training process, the parameters of the pre-training model are kept unchanged, and only the low-rank matrix and the mask parameters are updated; performing model evaluation after training is completed, and checking whether the model after knowledge distillation meets performance requirements or not; if the evaluation result is not satisfied, returning to the training step for further optimization; a low-rank matrix and mask parameters are fixed, and the parameters are fused into an original model according to masks. The model after knowledge distillation can obtain the capacity of a large model in a small size, and is more suitable for being deployed in a vehicle-mounted system with limited resources so as to provide efficient calculation performance.
Owner:BEIJING AUTOMOBILE WORKS CO LTD

PCB component defect detection method, system, equipment and medium

The invention discloses a PCB component defect detection method, system and device and a medium. The detection method comprises the steps that a large number of PCB component image data with labels are acquired to construct a training set and a test set; training a combined convolution and visual Transform network to detect the PCB component image data to obtain the type information and credibility of the image data; performing contour extraction on the PCB component image to obtain Mark point information; obtaining a character area of the PCB component image and obtaining character information of the image by using an OCR character recognition network; comparing the acquired type information, Mark point information and OCR character information with a standard PCB component, and judging whether the component has defects or not; the problems of low detection efficiency, poor detection precision and the like caused by manual detection can be solved, and high-efficiency and high-speed real-time detection can be completed under the condition of non-contact detection; in addition, the invention also comprises the system, the equipment and the medium in the method.
Owner:XI AN JIAOTONG UNIV

Improved sewer internal defect detection method and system based on YOLOv11

The invention provides a sewer internal defect detection method and system based on YOLOv11 improvement, and the method achieves the dynamic weight fusion of multi-level features through a self-adaptive feature fusion module, and improves the detection capability of a model for small targets and complex backgrounds. The selectivity of channel characteristics is improved through an SENet attention mechanism, and background interference is reduced; and introducing an MPDIOU loss function to improve bounding box regression loss. The performance of the improved model in detection tasks of sewer defects such as cracks, obstacles, corrosion, deformation and the like is remarkably superior to that of original YOLOv11, and particularly key indexes such as mAP50 (average precision), Box precision (P), recall rate (R) and the like are remarkably improved. According to the invention, efficient and accurate detection of multiple types of defects in the sewer can be realized, and a better solution is provided for maintenance and management of the sewer.
Owner:NANJING UNIV OF INFORMATION SCI & TECH

Light-weight metal plate surface defect detection method based on improved YOLOv5n

The invention discloses a lightweight metal plate surface defect detection method based on improved YOLOv5n, and relates to the field of machine vision and target detection, and the method comprises the following steps: S1, image preprocessing; s2, inputting the picture into the model; s3, model training; s4, data obtained after model training are analyzed and judged, the model is adjusted and optimized, and a final result is marked on an original image to serve as output of a detection result; and an improved YOLOv5n network model is adopted to carry out identification detection on the picture. The method has the advantages that the ADown module is introduced into the trunk network and the neck network to replace a CBS module in the original down-sampling process, the calculation efficiency is improved while the model parameter quantity is reduced, the deformable attention Transform is fused into the Bottleneck structure of the neck network C3 module, more feature information is obtained while the high reasoning speed is kept, and the method has the advantages of being high in reasoning speed, high in reasoning speed and high in reasoning speed. The method adapts to diversified defect characteristics of the surface of the metal plate, can meet the requirement for high detection precision, and meanwhile reduces the calculation amount of the model.
Owner:JIANGSU UNIV OF TECH

Large language model weight quantization method and system based on optimal quantization coding and application

The invention discloses a large language model weight quantization method based on optimal quantization coding, and the method comprises the steps: partitioning, normalization, quantization coding generation, quantization and inverse quantization, and achieves the weight quantization of a large language model, and comprises the following steps: 1, dividing a to-be-quantized weight matrix into a plurality of data blocks; step 2, carrying out normalization processing on elements in each data block; step 3, according to the normalized weight matrix value distribution characteristics, generating a quantization target code by adopting a clustering method; step 4, mapping elements in each data block after normalization processing in the step 2 to the index value of the quantization target code in the step 3 to realize element quantization; and 5, searching a floating-point number corresponding to the index value in the quantization target code, and multiplying the floating-point number with the normalized constant of the data block to obtain an inverse quantization floating-point number. The invention further discloses a system and application for implementing the method, and the system and the application have wide application scenes.
Owner:SHANGHAI QUSU CHAOWEI TECHNOLOGY CO LTD

Large model reasoning acceleration method and device and medium

PendingCN119204222AImprove parallel processing capabilitiesImprove inference speedInference methodsPhysical realisationData setAlgorithm
The invention discloses a large model reasoning acceleration method and device and a medium, and the method comprises the steps: constructing a training set and a test set for training a large model according to scene data; expanding a prediction head in the large model to obtain an expanded target large model, and freezing original prediction head parameters in the large model so as to update the prediction head parameters corresponding to the expanded prediction head in the training process of the target large model; training and evaluating the target large model through the training set and the test set to obtain a trained target large model; constructing a calibration data set formed by partial scene data, and inputting the calibration data set into the target large model for forward reasoning to obtain a calibration matrix corresponding to each linear layer; and determining an input matrix and an output matrix corresponding to the linear layer according to the calibration matrix, and quantizing a weight corresponding to the linear layer based on the input matrix to obtain a quantized model weight.
Owner:SHANDONG INSPUR SCI RES INST CO LTD

Lightweight bearing surface defect detection method and system based on improved YOLOv10n

The invention discloses a lightweight bearing surface defect detection method and system based on improved YOLOv10n. The method comprises the following steps: acquiring a bearing surface defect sample data set; the method comprises the following steps: improving a YOLOv10 network structure to obtain an FM-NET algorithm model; training an FM-NET algorithm model based on the bearing surface defect sample data set to obtain a target FM-NET algorithm model; and performing defect detection on a to-be-detected bearing image according to the target FM-NET algorithm model, and determining a defect position and a corresponding defect category in the bearing image. According to the method, the multi-scale feature extraction capability and the model expression efficiency are remarkably improved, the robustness of the model in a complex scene is enhanced, the reasoning speed is remarkably improved on the premise of keeping high precision, and therefore good balance between the precision and the efficiency is achieved.
Owner:YANCHENG INST OF TECH +1

Virtual face generation method and system, electronic equipment and storage medium

The invention provides a virtual face generation method and system, electronic equipment and a storage medium, and the method comprises the steps: inputting a to-be-processed face image, and audio features and face key point features corresponding to the to-be-processed face image into a coding module in a face generation model, and obtaining image coding features and lip motion coding features; inputting the image coding features and the lip motion coding features into a feature fusion module in a face generation model to obtain fusion features; inputting the fusion features into a decoding module in a face generation model to obtain a virtual face image; wherein the face generation model is obtained by training an initial image generation model, the parameter scale value of which is smaller than a preset value, based on a sample real face image, a mask face image corresponding to the sample real face image, and sample audio features and sample face key point features corresponding to the mask face image. According to the invention, the virtual face generation quality is guaranteed, and the virtual face generation efficiency is improved.
Owner:IFLYTEK CO LTD

Bird re-identification method and device, electronic equipment and computer storage medium

The invention relates to the technical field of artificial intelligence, and discloses a bird re-identification method, which comprises the following steps: obtaining an original bird image, and preprocessing the original bird image to obtain a preprocessed image; detecting whether birds exist in the pre-processed image, performing feature extraction on the pre-processed image when detecting that the birds exist in the pre-processed image to obtain a basic feature map, obtaining a preset two-stage recognition model, and performing fine-grained recognition on the basic feature map based on the two-stage recognition model to obtain bird information in the original bird image; and when it is detected that non-bird animals exist in the preprocessed image, obtaining a source bird watching device of the original bird image, and repelling the non-bird animals in the visual field based on the source bird watching device. The invention further discloses a bird re-recognition device, electronic equipment and a computer storage medium. According to the invention, fine-grained recognition of birds is realized, and the accuracy and precision of bird recognition are improved.
Owner:SHENZHEN FANYING ZHILIAN TECH CO LTD

Remote sensing target detection method and device based on multi-scale and efficient attention mechanism

The invention relates to a remote sensing target detection method and device based on multiple scales and an efficient attention mechanism, and the method comprises the steps: obtaining a target remote sensing image, inputting the target remote sensing image into a target detection model, and obtaining a target detection result; wherein the target detection model comprises the following steps: replacing a 3 * 3 common convolution in the sixth network layer ELANBlock of the YOLOv7 by using a 3 * 3 depth separable convolution; replacing a 3 * 3 common convolution in the ELANBlock of the eighth network layer of the YOLOv7 by using a 5 * 5 depth separable convolution; using a 7 * 7 depth separable convolution to replace a 3 * 3 common convolution in the ELANBlock of the tenth network layer of the YOLOv7; an MIA-M module based on multi-scale efficient attention is used for replacing the 17th network layer ELANBlock and the 23rd network layer ELANBlock of the YOLOv7; an MIA-S module based on multi-scale efficient attention is used for replacing the ELANBlock of the 21st network layer of the YOLOv7; and replacing the ELANBlock of the 25th network layer of the YOLOv7 with a multi-scale efficient attention MIA-L module to obtain a target detection model.
Owner:CHONGQING UNIV OF POSTS & TELECOMM

Table question and answer method, apparatus and device, readable storage medium and program product

The invention relates to a table question and answer method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring a form file and question data uploaded by a target user; analyzing table information in the table file to obtain a candidate label set; based on the question data and the candidate tag set, determining a spoken language understanding result corresponding to the question data, wherein the spoken language understanding result comprises an intention and a keyword corresponding to the question data; performing reverse verification on the spoken language understanding result according to the question data to obtain a verification result; under the condition that the verification result indicates that the spoken language understanding result passes verification, a cue word is constructed according to table information, question data and the spoken language understanding result in the table file; calling the large language model according to the cue word for question-answering to obtain a question-answering result. By adopting the method, a large-data-volume table can be processed, and meanwhile, the accuracy and the stability of large-model table question answering can be improved.
Owner:CHINA TELECOM CLOUD TECH CO LTD

Chinese named entity recognition method based on hierarchical label enhancement contrast learning

The invention discloses a Chinese named entity recognition method based on hierarchical label enhancement contrast learning. The method comprises the steps that 1, labels and characters are coded through a pre-training language model; 2, deep fusion is carried out on coding results through a cross-modal Transform network, and enhanced features are generated; 3, maximizing the similarity between the matched sentences and labels through a sentence level contrast learning method; 4, minimizing the difference between the character representation and the label enhancement representation through a lexical-level contrast learning method; 5, coding and decoding are carried out by using BiLSTM and CRF, and a final entity category prediction sequence is generated; and 6, constructing a total loss function and training the network. The Chinese named entity recognition method effectively improves the accuracy and generalization ability of Chinese named entity recognition, especially has excellent performance when processing complex scenes such as polysemy words and homosynonyms, does not need to depend on a dictionary, is high in reasoning speed, and is suitable for practical application scenes of large-scale text processing.
Owner:HEFEI UNIV OF TECH

Robot inspection defect detection method based on cross attention and multi-dimensional measurement

The invention discloses a robot inspection defect detection method based on cross attention and multi-dimensional measurement, and the method comprises the steps: designing a small sample target detection model, collecting a dam image through the autonomous inspection of a robot, and extracting a defect region in the image through a cross convolution attention method, and the defect areas are classified through multi-dimensional measurement learning, so that high-precision and high-efficiency defect detection work is realized. According to the method, a cross convolution module and a multi-dimensional measurement learning module are designed for solving the problem of disastrous forgetting and the problem that the quantity of model parameters is too large in an existing small sample target detection method, so that disastrous forgetting of the model is effectively relieved, and the model parameters are greatly reduced. In combination with the robot autonomous inspection technology, the method significantly improves the recognition efficiency and accuracy of the dam concrete surface defects, and realizes the intelligence and automation of dam safety monitoring.
Owner:HOHAI UNIV +1

Image restoration method and system based on gated convolution and fused gated attention mechanism

The invention discloses an image restoration method and system based on gated convolution and a fused gated attention mechanism, and belongs to the field of image super-resolution restoration frameworks, and the method comprises the steps: taking a preprocessed image as the input of an image restoration model, carrying out the feature extraction and dimension raising, and obtaining a feature map; the image features are extracted by fusing the attention and self-attention mechanism of the convolutional neural network type, and iterative optimization is carried out by using the loss function, so that the local and global features of the image can be processed more effectively, the image reconstruction quality is improved, and the authenticity and detail richness of the high-resolution image are ensured at the same time.
Owner:QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES)

Self-supervised image anomaly detection method based on patch aggregation and discrimination

The invention provides a self-supervised image anomaly detection method based on patch aggregation and discrimination, and belongs to the technical field of image anomaly detection. The method mainly comprises the steps of abnormal sample synthesis, depth feature extraction, patch aggregation, rapid attention feature fusion, feature correction and patch discrimination. According to the method, firstly, abnormal sample synthesis is carried out according to normal samples acquired from an image anomaly detection data set, then deep feature extraction is carried out on the normal samples and the abnormal samples, patch aggregation is carried out on extracted feature maps, the feature maps subjected to patch aggregation are fused by using rapid attention feature fusion, and the feature maps subjected to patch aggregation are subjected to feature fusion. And performing feature correction and patch discrimination on the fused feature map, integrating discrimination results of each scale, and performing up-sampling to obtain an input sample size so as to obtain a pixel-level discrimination result. According to the method, the problem that an artificially synthesized abnormal sample is not real enough in a conventional image anomaly detection method based on self-supervised learning is relieved, and meanwhile, the reasoning speed and accuracy are well balanced.
Owner:BOHAI UNIV

Real-time semantic segmentation method and system for multi-shape pyramid in traffic scene

The invention relates to the technical field of scene recognition, in particular to a real-time semantic segmentation method and system for a multi-shape pyramid in a traffic scene, and the method comprises the steps: initializing an original to-be-segmented image, and obtaining an initial segmentation feature map; channel splicing is carried out on the initial segmentation feature map and an output feature map obtained after first downsampling is carried out on the original to-be-segmented image, downsampling is carried out after a fused feature map is obtained, detail information is extracted, and a detail information feature map is obtained; down-sampling the detail information feature map and extracting semantic information; the method comprises the following steps: extracting transverse and longitudinal receptive field information of a semantic information feature map, performing expansion convolution to obtain feature maps of different receptive fields, obtaining a multi-scale and multi-shape context information feature map, performing up-sampling, performing point-by-point convolution dimension reduction on a detail information feature map, and performing channel splicing on the two features to obtain a fusion feature map containing semantic and detail information. And adjusting the channel number into a prediction category number, and normalizing to obtain a prediction segmentation result. The method has good segmentation precision and reasoning speed.
Owner:GUANGZHOU UNIVERSITY

Lightweight crack penetration degree detection method based on ground penetrating radar and Swin Transform optimization

The invention discloses a lightweight crack penetration degree detection method based on ground penetrating radar and Swin Transform optimization, and the method comprises the following steps: 1, collecting ground penetrating radar data, and carrying out the preprocessing and feature extraction; step 2, constructing a multi-feature fusion data set; step 3, designing a two-stage through crack detection model based on Swin Transform optimization; 4, a first-stage through crack detection model optimized on the basis of Swin Transformer-YOLOv8 is designed, and a second-stage through crack detection model is designed on the basis of Swin Transformer-YOLOv8 optimization; and step 5, training and predicting the data of the training set and the test set according to the model architecture in the step 3 and the step 4. According to the invention, the accuracy of crack penetration degree detection is improved, and a scientific basis is provided for road maintenance; the detection speed is greatly improved, and the requirement of large-area rapid detection is met; the calculation complexity is reduced, the equipment performance pressure is reduced, and the detection efficiency is improved; the generalization ability of the model is enhanced, and the method can adapt to different road conditions; and the development of an intelligent road detection technology is promoted.
Owner:CHANGAN UNIV

Vision and language fused multi-modal large model system

The invention discloses a vision and language fused multi-modal large model system, which comprises a vision encoder fused with DINOv2 and SigLIP and used for collecting low-level spatial attributes and semantic attributes; and the multi-layer perceptron projector is used for mapping the visual features to a language embedding space and a Mama backbone model network based on a state space model. Compared with a multi-modal large language model which depends on a Transform network as a basic model, the large model system disclosed by the invention has the advantage that indexes such as reasoning speed and accuracy are improved.
Owner:EAST CHINA NORMAL UNIV