Patents
Literature
178 results about "Goal recognition" patented technology
Efficacy Topic
Property
Owner
Technical Advancement
Application Domain
Technology Topic
Technology Field Word
Patent Country/Region
Patent Type
Patent Status
Application Year
Inventor
A deep learning method for plate defect detection based on dynamic receptive field and adaptive fusion
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.
Owner:SHANGHAI INST OF TECH +1
An intelligent panoramic perception and target tracking system based on cloud-edge cooperation
This invention discloses a cloud-edge-device collaborative intelligent panoramic perception and target tracking system, including an image processing layer comprising an NPU acceleration module and an image stitching module. The NPU acceleration module is used to extract multi-scale defect features from the panoramic video data using a lightweight convolutional neural network and output feature vectors. This invention achieves real-time edge processing of multiple video streams from the data acquisition layer via the NPU acceleration module and the key-value binding-based image stitching module at the image processing layer. This enables precise alignment and seamless fusion of images from multiple cameras, effectively solving the problems of poor stitching quality caused by large image distortion and asynchronous data frames in existing technologies. Furthermore, the system utilizes YOLO11m as the detection engine and an incremental learning interface at the server layer to perform high-precision target recognition and continuous cross-lens tracking of the panoramic image stream, allowing the system to dynamically expand its recognition capabilities and achieve low-latency intelligent perception.
Owner:NINGBO XINGBOYUAN INTELLIGENT TECHNOLOGY CO LTD
Voice instruction processing method and device, electronic equipment and storage medium
The present disclosure provides a voice instruction processing method and device, electronic equipment and a storage medium, relates to the technical field of artificial intelligence, in particular to the technical field of voice interaction. The specific implementation scheme of the voice instruction processing method is as follows: performing multiple path recognitions on a to-be-processed voice instruction to obtain multiple recognition results, wherein the multiple path recognitions include first operation intention recognition of performing an operation on a graphical interface element of a voice interaction device and second operation intention recognition of performing an operation on the voice interaction device; obtaining a target recognition result from the multiple recognition results; and performing the to-be-processed voice instruction according to the target recognition result.
Owner:APOLLO INTELLIGENT CONNECTIVITY (BEIJING) TECH CO LTD +1
Target identification method and device, and vehicle-road-cloud integrated test system
Owner:CHONGQING CHANGAN AUTOMOBILE CO LTD
Construction method, construction method and equipment of evidence photograph recognition model based on AI national investigation
The application discloses an AI-based land investigation evidence photograph identification model construction method, an identification method and equipment, and the method comprises the following steps: acquiring a training data set; using the preprocessed historical evidence photographs to train the constructed initial identification model, obtaining the target identification model corresponding to each region and each period, and the target identification model is used for deploying on each terminal equipment. Through the quality detection and classification identification of the collected evidence photographs by the target identification model deployed on the terminal equipment, the problem that the business level of field personnel is uneven in the land investigation evidence work, which leads to inaccurate land class identification in the evidence photograph shooting process, and the real-time judgment of the land class identification and the evidence photograph shooting quality in the evidence shooting process by using the AI technology is solved, so that the shooting evidence photograph quality is improved, the overall land change investigation result is improved, and the subsequent land investigation evidence work is effectively guided.
Owner:曹扬 +3
A video target recognition method based on multi-model hot switching
The application discloses a video target recognition method based on multi-model hot switching, and belongs to the technical field of computer vision and video processing. The method comprises an arbitration module, a switching control module and a feature adaptation and buffer module. The arbitration module generates a switching preparation signal by extracting multi-dimensional indexes such as optical flow mean, local variance, target density and scene confidence in real time through a lightweight channel independent of main reasoning. The switching control module performs atomic replacement of a computation graph pointer in a vertical blanking period, and realizes millisecond-level hot switching with zero frame loss by combining an asynchronous pre-copy and a chasing mechanism of a double buffer. The feature adaptation and buffer module solves tensor shape mismatch between heterogeneous models through a pre-compiled adaptation layer. The application also provides optimization schemes such as multi-index nonlinear fusion decision, zero-copy memory management, local slice focus reasoning and edge-cloud hierarchical unloading, significantly reduces switching delay, guarantees continuous recognition of a video stream, and is suitable for edge computing scenes with limited resources.
Owner:SICHUAN BAICHUAN SIWEI INFORMATION TECH CO LTD
Target comprehensive identification method, device and equipment based on improved d-s evidence theory
The application provides a target comprehensive identification method and device based on improved D-S evidence theory and an electronic device, belongs to the technical field of intelligent control, and aims to solve the problem that current target identification methods rely on single sensors or single feature information, are difficult to adapt to target identification requirements in complex environments, and result in low identification efficiency and accuracy. According to the source of detection information and the information type contained in the detection information, double weight values can be given to first target detection information, second target detection information, third target detection information and fourth target detection information, the flexibility of the first target detection information, the second target detection information, the third target detection information and the fourth target detection information can be improved to a certain extent, the target detection function value of the target area can be calculated by using the D-S evidence theory method with weight, and then the first missile identification result is obtained, the fusion accuracy of multi-source data is improved to a certain extent, and the accuracy and reliability of the first missile identification result are improved.
Owner:AIR FORCE UNIV PLA
A method for generating an adversarial sample of a SAR image and application
The application discloses a kind of SAR image's adversarial sample generation method and application, belong to synthetic aperture radar automatic target recognition security technical field;The present application is based on the average value of using integral integrated gradient, by randomly selecting part of model in integrated model multiple times, and calculating the difference of the temporary adversarial sample gradient generated in inner and outer loop, to correct the outer loop integrated gradient, so that its update direction is more accurate, can generate high-quality adversarial samples in real black box scene, and then can improve the success rate of attack in black box attack, and then affect the accuracy of security detection, in addition, it can also be used to reinforce depth learning model, such as in the process of training model, by adding the generated adversarial sample in data set can further construct better model with robustness, and then can effectively defend the attack of adversarial sample.
Owner:HUAZHONG UNIV OF SCI & TECH
A multi-modal target detection and recognition method based on image processing
The present application relates to the field of target identification, and more particularly to a multi-modal target detection and identification method based on image processing, comprising: determining and extracting strategy based on perspective richness and feature stability; when performing uniform extraction, determining whether to adjust the extraction quantity based on perspective satisfaction; when performing compensation extraction, determining the compensation paragraph based on the regional parameter representation value; for the reference image obtained by extraction, determining whether to perform feature point optimization based on the feature anchor representation value and the spatial coverage; in the feature point optimization process, determining the point optimization strategy based on the evaluation balance; determining the sub-region category based on the regional prominence, and determining whether to adjust the feature region area based on the one-class quantity and the one-class aggregation degree. The present application effectively improves the multi-modal re-identification effect.
Owner:BEIJING TOPMOO TECH
Industrial target recognition and accurate positioning method based on improved Hough algorithm
The application belongs to the technical field of industrial system information, and provides an industrial target recognition and accurate positioning method based on an improved Hough algorithm, which comprises the following steps: adjusting parameters of an industrial camera to obtain clear image data of a workpiece, and pre-processing the image data; performing edge detection on the pre-processed image based on a morphological processing algorithm and a Canny edge detection algorithm, obtaining edge information, and judging whether there is edge loss; for the edge loss condition, performing target recognition and positioning based on an improved Hough algorithm or a template matching algorithm, and obtaining the center point coordinates of the target object; the application can more accurately and effectively detect the features of the target boundary, accurately deduce the boundary loss condition, and then realize accurate positioning of the detection target.
Owner:KUOJING INTELLIGENT MANUFACTURING (TIANJIN) HIGH-TECH CO LTD +2
A three-dimensional point cloud intrusion detection method and device based on deep learning
ActiveCN121033758Bachieve aggregationImprove feature expression qualityCharacter and pattern recognitionPattern recognitionVoxel
This invention discloses a method and apparatus for intrusion detection based on deep learning-based 3D point cloud. The method acquires a point cloud of the scene to be tested; it then performs target recognition on the point cloud using a personnel target detection network to obtain personnel target detection information. The personnel target detection network is constructed based on the Voxel R-CNN framework, and the feature set of the voxel grid is generated based on the position information of the points in the point cloud contained within the voxel grid and the point cloud features. Based on the detection information and the position information of dangerous areas in the scene to be tested, it determines whether an intrusion event has occurred. This invention compensates for the shortcomings of MeanVFE in feature utilization by reconstructing the voxel grid feature set generation method through the personnel target detection network, calculating the offset, and concatenating it with other features. It also achieves the aggregation of local and global information, utilizing feature information at different levels to improve feature expression quality, thus significantly improving detection accuracy.
Owner:XIAN UNIV OF SCI & TECH
Reinforcement learning tuning method and apparatus for end-to-end object detection algorithms
ActiveCN118761452BCharacter and pattern recognitionBiological modelsGoal recognitionReinforcement learning algorithm
The application relates to a reinforcement learning optimization method and device for an end-to-end target detection algorithm. The method comprises the following steps: obtaining image data to be processed, extracting image data features of the image data to be processed by using a hyperparameter optimization model, selecting corresponding hyperparameters in combination with corresponding task information, performing algorithm optimization on an end-to-end algorithm by using the hyperparameters, obtaining an optimized end-to-end algorithm, performing target recognition by using the optimized end-to-end algorithm, evaluating a target detection recognition result, selecting corresponding rewards, updating the hyperparameter optimization model based on the rewards in combination with a reinforcement learning algorithm gradient, and obtaining an optimized parameter optimization model. The method of the application does not need manual design for optimization, does not need to set a fixed manual detection threshold, greatly improves the algorithm optimization efficiency, and significantly improves the generalization of the algorithm.
Owner:BAIYANG FUTURE (BEIJING) TECH CO LTD
Dynamic target recognition method and system based on low-illumination environment
The application belongs to the technical field of target identification, and provides a dynamic target identification method and system based on a low-illumination environment. An illumination component and a reflection component are separated frame by frame through a Retinex decomposition network, and an illumination distribution map is generated based on the illumination component. The convolution kernel weight of an edge branch is modulated based on the illumination distribution map, and an edge feature map and a semantic feature map are extracted. A candidate region and a corresponding uncertainty distribution are generated based on the illumination component and the reflection component to determine a fusion weight and complete weighted fusion, thereby improving the integrity and recognition degree of the fusion feature map. Target detection and confidence determination of the candidate region are performed based on the fusion feature map, local secondary enhancement is performed on a low-confidence region, and a target region is output after inter-frame residual fusion based on the fusion feature map of adjacent frames, thereby improving the stability and environmental adaptation capability of cross-frame dynamic target identification.
Owner:NANJING SHIYUN INFORMATION TECH CO LTD
Model training apparatus and recognition apparatus for target recognition in medical images
The present disclosure describes a model training device and a recognition device for target recognition in medical images. The model training device comprises an acquisition module, a construction module and a training module; the acquisition module is configured to acquire a medical image as a training sample and a labeled region corresponding to a target in the training sample; the construction module is configured to determine a target region of the target within the labeled region, and construct a training set using the training sample, the labeled region and the target region, pixels within the target region in the training sample being determined to belong to the target; and the training module is configured to train a to-be-trained model based on the training set, and optimize the to-be-trained model using a training loss function to obtain a trained model, wherein in the training loss function, a spatial weight is used to reduce the negative impact of pixels in a first region in the training sample on the to-be-trained model, the first region being a region outside the target region of the target within the labeled region in the training sample. Thus, small targets can be effectively recognized.
Owner:SHENZHEN SIBRIGHT TECH CO LTD
A collision target recognition method and related apparatus
The application discloses a collision target identification method and related device, and relates to the field of automobile control. The method comprises the following steps: after predicting a self-vehicle track and a target predicted track, the position of the target object relative to the self-vehicle and the yaw angular velocity of the target object are determined, the target predicted track is corrected based on the yaw angular velocity, and then the collision target is identified based on the self-vehicle predicted track and the corrected target predicted track. In the process, the position of the target object relative to the self-vehicle and the yaw angular velocity of the target object are used as reference quantities for determining whether the target predicted track needs to be corrected, and the track of the target object is no longer kept in uniform straight line or uniform variable speed straight line motion, so that the collision object identification system can accurately predict and identify the collision target.
Owner:BEIJING CO WHEELS TECH CO LTD
A fusion of track and image collaborative target recognition method and system
The application relates to the technical field of target identification, and discloses a kind of fusion track and image collaborative target identification method and system, the method comprises: synchronously obtaining track data and image data, and track data and image data are time-space registered, and collaborative data is obtained;Based on collaborative data, track data is time-series correlation fitting, and target track is obtained;Based on collaborative data, image data is multi-scale feature fusion, and image fusion atlas is obtained;Target track and image fusion atlas are bidirectional intercommunication correlation, and track atlas pair is obtained;The behavior semantics of track atlas pair is interpreted, and behavior intention is obtained, and the situation representation between behavior intention is evaluated, and situation representation is obtained;Based on situation representation, track atlas pair is structured analysis, and target identification result is obtained;The application can improve the accuracy of collaborative target identification.
Owner:ZHEJIANG LANJIAN DEFENSE TECH CO LTD
A remote sensing image segmentation method based on integrated cross-attention mechanism
This invention discloses a remote sensing image segmentation method based on an integrated cross-attention mechanism. The method employs a deep residual network ResNet101 to extract features from RGB remote sensing images, obtaining an initial feature map. A cross-attention mechanism is introduced, using the original image and the initial feature map as input, to mine the regional associations within the RGB remote sensing image, enhance key region features for target recognition, and perform feature enhancement on the initial feature map, outputting a focused feature map. A deep learning model, DeepLabV3, is used to extract multi-scale features from the focused feature map. The Awfully Hollow Spatial Pyramid Pooling (ASPP) module captures contextual features under different receptive fields and fuses them with low-level detail features from ResNet101, resulting in a final feature map with rich context and clear boundaries, outputting the semantic segmentation result. This invention improves the segmentation capability of complex scenes through the cross-attention mechanism, and by utilizing image space and contextual information, it can effectively improve the segmentation accuracy of complex urban and rural features.
Owner:HANGZHOU NORMAL UNIVERSITY
Multimodal domain incremental railway fastener defect detection method and system in bad weather
This invention provides a multimodal incremental railway fastener defect detection method and system under severe weather conditions, belonging to the field of target recognition technology. Step 1: Sequentially construct training and validation sets for multimodal railway fasteners under rain, snow, fog, strong light, low light, and normal weather conditions; construct a global test set containing multimodal railway fastener data for all six weather conditions. Step 2: Construct a full-cycle closed-loop mechanism for multimodal target detection to achieve continuous learning in continuously evolving scenarios. Step 3: Propose a dual-space replay screening strategy for meta-features to extract challenging and representative high-value samples for inclusion in the replay pool. Step 4: Propose a wavelet-guided non-adversarial detail injection module to resolve the adversarial conflict between noise removal and style preservation in incremental learning. This invention can continuously adapt to various severe weather changes in railway scenarios, effectively alleviate the problem of catastrophic forgetting, and achieve accurate detection of fastener defects under various severe weather conditions.
Owner:BEIJING JIAOTONG UNIV
Pulse neural network continuous learning target recognition method based on space-time information fusion
PendingCN122289806Amitigation of catastrophic forgettingStable prior knowledge across tasksTime informationFeature extraction
This invention belongs to the field of image recognition technology, specifically relating to a continuous learning target recognition method based on spatiotemporal information fusion using a spiking neural network. It comprises two learning modules: a fast learner and a slow learner. The slow learner acquires general features that are invariant to input changes but sensitive to semantic representation through self-supervised learning. In the fast learner, whenever a new task arrives, the processor freezes the trained feature extraction modules and adds new feature extraction modules to learn features of the new category. These feature extraction modules gradually aggregate to form a joint feature representation. By fusing general representations with task-specific features across spatial and temporal scales, this method effectively mitigates task confusion and catastrophic forgetting, improving target recognition accuracy.
Owner:ZHEJIANG UNIV
Object recognition and object information acquisition system and method using a telescope
The object recognition and object information acquisition system and method using a telescope according to the embodiment provides a recognition alarm when a specific object (an opponent's vessel) is recognized while monitoring through a telescope in a naval environment. In addition, in the embodiment, information regarding the object is acquired through artificial intelligence. In the embodiment, the information regarding the object may include the size, year of manufacture, purpose, output, and speed of the vessel. Furthermore, in the embodiment, the distance to the object vessel is predicted by using a calculation formula that combines the size of the captured object based on the information and the actual size of the vessel obtained through AI. Additionally, in the embodiment, an object is recognized in an image captured by a telescope, and information regarding the recognized object is acquired through image search and artificial intelligence.
Owner:BAEKSAN SCIENCE CO LTD
Text recognition method, apparatus, device, medium, and product
The application discloses a text recognition method, device, equipment, medium and product. The text recognition method comprises the following steps: setting a target recognition template for a target type of certificate; obtaining a target certificate image corresponding to the target certificate, performing text recognition on the target certificate image, and obtaining an initial recognition result; determining a plurality of pairs of target text regions and target anchor point regions matched with text information; determining mapping region position information corresponding to a mapping region of a to-be-recognized region in the target certificate image according to target region position information; determining a first text region intersecting with the mapping region from a plurality of text regions; obtaining first text information corresponding to the first text region from the initial recognition result, and outputting a recognition result corresponding to the target certificate image according to the first text information. According to the embodiment of the application, the text recognition of different types of certificates can be performed based on a general recognition model, and a large amount of time and resources are saved.
Owner:CHINA CONSTRUCTION BANK +1
A method and apparatus for identifying stable targets based on visual feature-based random partitioning and decorrelation.
This invention proposes a stable target recognition method and apparatus based on visual feature random partitioning and decorrelation. The method includes: acquiring a visual feature set of a training dataset consisting of industrial images and their defect labels, and performing multiple random partitions; after each partition, all dimensions of visual features are divided into two mutually exclusive feature segments; under each feature partitioning scheme, using the training dataset, learning a set of sample weights to decorrelate the two feature segments in a weighted distribution; weighting the training data using the sample weights, and using the weighted training data to train a base recognition model; inputting the image to be recognized into the trained base recognition model to obtain the corresponding recognition result; and integrating the recognition results from all models to obtain the final stable recognition result. This invention can effectively alleviate the environmental bias problem in industrial production lines and improve the stability and generalization performance of defect detection models with limited training data.
Owner:TSINGHUA UNIVERSITY
Target recognition method and device, electronic equipment, storage medium and program product
The present disclosure relates to a target identification method and device, electronic equipment, storage medium and program product, comprising: obtaining an original image dataset, dividing a general training set and an expert sub-training set, and comparing the data leakage rate; training a general model and an expert model, adjusting the expert training strategy according to the leakage rate; evaluating the performance index and the generalization gap of the model, calculating the adaptive fusion weight combined with the leakage rate, and weighting and fusing the initial fusion model; fine-tuning the final fusion model through a low learning rate; inputting the to-be-identified image into the final model to obtain a candidate box, eliminating false positives through multi-modal large model semantic discrimination, and obtaining the final identification result. Among them, through the hierarchical architecture of the general sub-model and the multiple expert sub-model, the differences existing in the same type of target are modeled, and after adaptive weight fusion and unified fine-tuning, the recall rate of the model is significantly improved in the scene with large intra-class differences, the omission rate is significantly reduced, and high-precision detection and stable output of heterogeneous clusters are realized.
Owner:CHINA ELECTRONICS CLOUD DIGITAL INTELLIGENCE TECH CO LTD
A method for autonomous decision making for a ship unmanned system
This invention relates to an autonomous decision-making method for unmanned ship systems. The method includes acquiring sensory data from the unmanned system, constructing a local expert knowledge base, building an autonomous decision-making thought chain, inputting image data into a small target detection model to obtain target category information and corresponding confidence levels, acquiring target recognition results, inputting the target recognition results and the thought chain into a multimodal large model, outputting a decision action, comparing and evaluating the output decision action with the corresponding decision actions in the expert knowledge base, obtaining the final output decision action, locating problems in erroneous decision actions, and iteratively correcting the problems. This decision-making method can improve the target recognition accuracy of various sensory data, streamline the reasoning process, improve the real-time performance of decision output, and ensure the reliability of decision actions. Simultaneously, it can automatically identify image recognition errors and decision output errors, perform iterative corrections, and enable the system to continuously improve recognition accuracy and decision rationality during operation.
Owner:JIANGNAN SHIPYARD (GRP) CO LTD
Space-based and air-based target cooperative identification method and device
Owner:TSINGHUA UNIVERSITY
A method and device for radar target recognition with azimuth enhancement under physical scattering constraints
PendingCN122265763AEnsure physical authenticityensure reliabilityWave based measurement systemsCharacter and pattern recognitionRadarCharacteristic space
The application discloses a radar target recognition method and device with azimuth enhancement under physical scattering constraints. The core is that a feature level azimuth expansion module (FAEM) is constructed, sparse view angle features are augmented in a feature space, and effective expansion of azimuth information is realized; meanwhile, a scattering center reconstruction constraint module (SCRCM) is innovatively introduced, physical authenticity and semantic fidelity of generated features are improved; in addition, multi-view perception enhancement modules (MVFM) are used to fuse generated multi-azimuth features, and intra-class compactness constraints are applied to optimize feature space structures, and the recognition ability of the model is further improved. The method realizes optimization from feature generation, physical constraints to semantic enhancement, and effectively improves the target recognition performance under sparse view angles.
Owner:BEIJING INST OF TECH
A rapid target detection and localization method, device and unmanned aerial vehicle system
This invention discloses a rapid target detection and localization method, apparatus, and unmanned aerial vehicle (UAV) system. The method includes: acquiring a two-dimensional scene image and scene depth information of a predetermined scene; processing the scene image using a target detection model to identify the target and its key features; solving for the target information and key feature information of the key features; obtaining the target's spatial three-dimensional attitude information using a target spatial mapping model based on the target information and key feature information; solving for the depth information of the target and key features relative to the UAV based on the scene depth information; and solving for the target's spatial pose information relative to the UAV based on the spatial three-dimensional attitude information and the depth information of the target and its key features relative to the UAV. This invention has the advantages of low computational complexity, high computational speed, and low resource consumption, and can improve the target recognition and localization efficiency of UAV systems and the mission endurance of UAVs.
Owner:CHINA NANHU ACAD OF ELECTRONICS & INFORMATION TECH