91 results about "Region proposal" patented technology

The invention belongs to the technical field of computer vision and image processing, and discloses a remote sensing small sample target detection method based on double-attention guided transfer learning, and the method comprises the steps: obtaining a remote sensing image data set, and carrying out the preprocessing; taking the preprocessed remote sensing image training set as input, constructing a basic detection model by using a ResNet-101 backbone network, a feature pyramid network and a content awareness upsampling and regional proposal network, and obtaining basic model parameters; basic model parameters are used as input, a DA-FSDET network is trained based on a content awareness strip pyramid and a deformable attention area proposal network, and the trained DA-FSDET network is used to acquire a category detection frame containing small sample categories and confidence. Through cascading and cooperative work of the content awareness stripe pyramid and the deformable attention area proposal network, the detection precision and robustness of the multi-scale target in the remote sensing image are effectively improved.

One embodiment of the present invention sets forth a technique for performing a labeling task. The technique includes determining one or more region proposals, wherein each region proposal included in the one or more region proposals includes estimates of one or more bounding boxes surrounding one or more objects in a plurality of video frames. The technique also includes performing one or more operations that execute a refinement stage of a machine learning model to produce one or more refined estimates of the one or more bounding boxes included in the one or more region proposals. The technique further includes outputting the one or more refined estimates as initial representations of the one more bounding boxes for subsequent annotation of the one or more bounding boxes by one or more users.

The invention discloses a remote sensing target detection method for an open set scene, and belongs to the technical field of open set target detection. The open set remote sensing target detection network introduces a cascade target positioning network, a background feature enhancement module and reciprocal point prototype learning on the basis of a Faster-RCNN model; the cascade target positioning network can effectively adapt to multi-scale features, calculates the quality of a candidate box through centrality loss, and replaces a foreground-background classification mode in a traditional region suggestion network, so that the recall rate of objects is increased, and the possibility that unknown objects are misjudged as backgrounds is reduced; the background feature enhancement module fuses various background region information so as to enhance the adaptability to a complex environment and reduce the interference of background noise on target recognition; according to reciprocal point prototype learning, a reciprocal point prototype of a known category and a background is constructed, and an embedded network metric learning strategy is combined, so that background information and the known category jointly participate in candidate box classification, and a known target, an unknown target and the background are effectively distinguished.

The invention discloses a tiny target detection method and system based on a hybrid enhancement network, and the method comprises the steps: taking a remote sensing image as the input of a target detection network, carrying out the preprocessing and multi-stage feature processing of the remote sensing image through a backbone network, extracting the multi-scale texture features and global context features, and carrying out the recognition of the features; obtaining fusion feature maps of different stages; inputting the fused feature map into a feature pyramid network and a region proposal network in sequence to generate a tiny target candidate box; extracting fixed-size features from the candidate box, inputting the fixed-size features into a mixed attention enhancement module, and learning context-enhanced tiny target feature representation; the prototype library is dynamically updated by using the prototype comparison loss of the prototype comparison learning module, the target detection network is optimized, tiny target classification and positioning are completed through a directional detection head, and a detection result is output; the method has the advantages of strong feature extraction capability and high robustness.

The invention discloses an improved Faster R-CNN underground drainage blow-off pipeline defect detection model, and particularly relates to the field of target detection, and the method comprises the steps: firstly constructing a defect image data set, extracting a defect-containing image from a pipeline detection image, and dividing the defect-containing image into a training set, a verification set and a test set according to a preset proportion after validity screening and defect category labeling; secondly, improving a detection model, replacing a traditional regional proposal module with an anchor-frame-free detection head, embedding an attention module in a high layer of a backbone network, and introducing a feature pyramid network to generate multi-scale features; then training the model, performing multi-graph splicing type data enhancement on a training set, training by adopting an optimizer and a two-stage loss function, and storing an optimal weight in combination with an early stop strategy; and finally evaluating the model, inputting the test set to obtain a detection result, counting related detection data, calculating a precision index, and comparing the precision index with a preset threshold to judge whether the model reaches the standard. The method is high in detection precision and high in adaptability, and meets the requirement for pipeline defect detection.

The invention discloses a cross-domain small sample target detection method based on dynamic information fusion, relates to the field of industrial defect detection and the like, and aims to solve the problems of inter-domain distribution difference, annotation scarcity and the like in a cross-domain small sample scene. The method specifically comprises the following steps: firstly, dividing a source / target data set, constructing a DIC-ViT model containing double branches, and sharing a DINOv2 backbone network to extract features; an anchor box is generated through a region proposal network, and sample division is optimized by combining dual criteria of an IoU value and a category center distance; then feature adaptive fusion is realized through a dynamic information coupling module, and a discriminative category prototype is generated through a contrast learning module; and finally, optimizing the model by using an objective function containing classification, regression and contrast loss. The method can improve the target detection robustness and precision, is excellent in performance in a multi-data-set test, and is suitable for a detection scene with scarce samples.

The application discloses a three-dimensional target detection method based on a multi-modal feature fusion of a Transform, and comprises the following steps: 1. collecting point cloud data by using a laser radar and sampling the point cloud data, and collecting image data by using a camera; 2. inputting the data collected by the laser radar and the camera into a multi-modal feature fusion RPN network based on a Transform to extract a region proposal frame; and 3. inputting the region proposal frame information into a refinement network to obtain a final prediction frame. The application can avoid the problems of false detection and missed detection in the target detection process based on the multi-modal feature fusion, so that the accuracy of the evaluation of the perceived environment can be ensured.

The invention relates to the technical field of computer vision and intelligent traffic, in particular to a road spilled object detection method and device based on a visual large model, and the method comprises the steps: firstly constructing multi-scale feature representation for an image, and obtaining a context enhancement feature map through a feature coding network containing a multi-head self-attention mechanism; generating a candidate region by a region suggestion network based on the feature map and giving a position parameter; performing instance separation processing in the candidate region to form instance-level candidate region features; and inputting the features into a classification sub-network and a boundary regression sub-network, and outputting a bounding box and a category label. According to the method, boundary continuity and positioning consistency are kept in close, overlapped and weak texture scenes, and the method is suitable for road safety monitoring and vehicle auxiliary identification.

The application discloses a kind of dual paradigm combined remote sensing image end-to-end fine-grained target detection method, it is related to computer vision and pattern recognition technical field, including: extracting multi-scale feature map by feature extraction network;Based on sparse directional proposal network, using one-to-one sparse matching strategy, generate no duplicate sparse directional region proposal and its corresponding proposal feature and coarse-grained target confidence;Based on query perception refinement head, using one-to-many dense matching strategy, obtain fine-grained classification probability and refined bounding box;In the training stage, introduce double auxiliary supervision head to provide additional dense supervision signal;In inference stage, adopt two-stage consensus scoring mechanism, finally obtain fine-grained target detection result.The method realizes the end-to-end remote sensing image fine-grained target detection without NMS dependence, through the cooperation of sparse and dense dual paradigm, effectively improves the positioning accuracy and fine-grained classification performance.

A method of performing target detection during autonomous driving, comprising: performing 3D target detection in a 3D target detection segment; uploading outputs of a plurality of sensors in communication with the 3D target detection segment into a plurality of point clouds; transmitting point cloud data of the plurality of point clouds to a region proposal network (RPN); independently performing 2D target detection in a 2D target detector, the 3D target detection being performed in parallel in the 3D target detection segment; and obtaining a given input image and simultaneously learning bounding box coordinates and class label probabilities in a 2D target detection network that operates to treat target detection as a regression problem.

A kind of construction method and device of depression recognition model based on deep network feature fusion, its method includes: data acquisition;Build the video dataset containing video regional level label;Respectively using C3D neural network and channel attention network to the whole video stream is reinforced feature extraction;The reinforced feature is used RPN to generate region proposal and produces region start mark and end mark and region expression class;All video frames are extracted to multiple scale features including local eccentricity features and global features;Weakly supervised classification is carried out to multiple scale features;The operation result of video stream level feature and picture level feature is classified by neural network, and depression recognition model is obtained.The present application effectively utilizes video label, takes into account global and local information, and combines the advantages of high accuracy and reduced label workload, can become a potential depression recognition auxiliary means, and can be widely applied in the field of automatic processing of mental illness.

The application provides a cross-domain small sample wideband signal detection and recognition method based on a visual basic model, and relates to the cross technical field of electromagnetic signal processing and computer vision. The application reduces the interference of background noise in the wideband time-frequency graph on the candidate area by performing double condition screening based on target confidence and aspect ratio features on the preliminary candidate frame, and combining the field of view expansion in the frequency axis direction, and improves the incomplete coverage problem of the slender signal candidate frame. By extracting multiple intermediate layer features and the final layer feature map of the visual basic model for fusion in the channel dimension, the problem that the local texture information in the wideband time-frequency graph is smoothed or covered in the layer-by-layer abstraction process is solved. The parameters of the visual segmentation model and the visual basic model are kept frozen, and only the small sample fine tuning of the front region proposal network is performed, so that the calculation overhead caused by full fine tuning of the visual basic model during cross-domain small sample adaptation is avoided, and the method is suitable for deployment on devices with limited computing resources.

This invention discloses a controlled knowledge transfer-based generalized small-sample target detection method and system for marine remote sensing. The method includes the following steps: extracting multi-scale features from remote sensing images to generate multi-scale feature representations; generating candidate region features from the multi-scale feature representations via a region proposal network; inputting the candidate region features into a parameter-independent base class feature extraction branch and a new class feature extraction branch to generate base class features and new class features, respectively; performing a two-level fusion process on the base class features and new class features to generate fused features; applying dual constraints to the fused features using knowledge distillation loss and mutual information regularization loss, whereby the knowledge distillation loss constrains the consistency between the fused features and the base class features, and the mutual information regularization loss reduces the redundant correlation between the new class features and the base features; and inputting the fused features into a classification branch and a regression branch to obtain the detection results.

The present disclosure provides a juvenile fish limb identification method based on a multi-scale cascaded perceptual convolutional neural network. The method includes the following steps: acquiring a video sequence of a juvenile fish, dividing a fish body into five non-overlapping parts, performing semantic annotation on the five non-overlapping parts, and taking the five non-overlapping parts as an input of the multi-scale cascaded perceptual convolutional neural network; and using a convolutional layer as a feature extractor, performing feature extraction on an input image containing the annotation of each limb, inputting extracted features into an Attention-region proposal network (RPN) structure, determining a category of each pixel, and generating a limb mask of each limb category using a multi-scale cascade method. According to the method, the limbs of the juvenile fish can be efficiently and accurately identified, and technical support is provided for posture quantification of the juvenile fish.

The present application relates to the technical field of detection data processing, and particularly relates to a textile fiber component analysis system and method based on machine learning; the direction gradient attention module provided in the embodiment has strong direction bias in feature representation before generating a candidate region proposal, which makes the region proposal network more easily locate the geometric center of the fiber according to the extension direction of the fiber, so as to generate a more compact and accurate candidate box, and lay a foundation for subsequent disentanglement of the adhered fiber. By focusing attention on several main directions related to the fiber, the direction gradient attention module naturally suppresses the edge response from random direction background noise, purifies the features, and improves the robustness of the model.

The invention relates to an open vocabulary target detection method based on a region proposal network and region correction, and the method comprises the steps: carrying out the feature extraction of a target image, and obtaining an initial feature map; inputting the initial feature map into a pre-constructed region proposal network to obtain a potential target region; based on the potential target region, guiding the initial feature map to perform region feature extraction; carrying out attention pooling processing on the extracted region features to obtain region embedding; performing text coding processing on categories in the detection list to obtain text embedding; and calculating the cosine similarity between the region embedding and the text embedding, and taking the category corresponding to the text embedding with the maximum cosine similarity with the region embedding as the category of the corresponding region. According to the invention, the accuracy and robustness of a target detection system in processing new types of targets can be improved, so that the high requirements of application scenes such as intelligent monitoring and automatic driving on target detection are met.

The invention discloses a 3D target detection method based on point cloud context structure information. The method comprises the following steps: collecting point cloud by using a data collector; voxelizing the point cloud to construct a 3D voxel representation; sending the 3D voxel representation into a backbone network to extract a multi-scale 3D voxel feature and a 2D BEV feature of the point cloud; sending the 2D BEV features into a regional proposal network to generate an initial proposal box; performing graded refining on the initial proposal box and the multi-scale 3D voxel features to generate an updated proposal box; and outputting a detection result according to the updated proposal box. According to the 3D target detection method, multi-scale point cloud context structure information can be effectively combined, the initial proposal box is iteratively optimized, and the detection performance is remarkably improved.

The application discloses a kind of feature fusion-based remote sensing image small target detection method and system, and the specific steps of the method are as follows: the remote sensing image to be detected is input into the network architecture mainly constituted by convolutional neural network and feature fusion network;Image will enter convolutional neural network and region proposal network after entering network architecture, region proposal network will cooperate with convolutional neural network to carry out pooling operation on image region of interest, while the deep feature of image region of interest is output to feature fusion network, image will obtain image region of interest after entering region proposal network, and the manual feature of image region of interest is output to feature fusion network by calculation;Deep feature and manual feature enter feature fusion network, and feature fusion network includes two feature conversion modules and a feature embedding module.

The application discloses a video target detection method based on category-aware feature aggregation, comprising: detecting a main stem ResNet-101 to perform feature extraction on a video frame to obtain high-level semantic information; using deformable convolution to perform sub-pixel level feature alignment, and then using a region proposal network RPN to generate a target candidate frame for each frame; using a candidate frame classification module to perform classification operation on the generated candidate frame, and subsequently performing feature aggregation only on candidate frames with the same class label; when performing category-aware feature aggregation, using a wide-range layer-by-layer progressive feature aggregation module to perform bidirectional and layer-by-layer progressive feature aggregation operation on the candidate frame level; using an inter-class relationship modeling module to model the spatial position relationship of different category targets on the same frame; and inputting the aggregated candidate frame features into a full connection layer to perform specific category discrimination and regression correction of the target frame position. The application fully utilizes the time sequence information and the spatial position information, thereby effectively improving the detection performance of the video target.

The application discloses an insulator defect detection method based on an improved neural network model and a storage medium, and the method comprises the following steps: acquiring an image of an insulator to be detected; performing defect detection on the image of the insulator to be detected by using a trained improved neural network model to obtain a defect detection result of the image of the insulator to be detected, wherein the model comprises a feature extraction module used for extracting features of an image, a region proposal module used for generating a candidate box, an interest domain pooling layer and an output module used for classification and regression. The application realizes multi-scale fusion of deep features and shallow features, enhances the weight of important features, improves the capture probability of a target and the detection accuracy of targets of different scales, simultaneously improves the target frame regression mechanism of the model and the scoring mechanism when the overlap degree of a predicted frame is high, reduces the probability that a predicted frame is mistakenly deleted, and can effectively improve the recognition accuracy and precision of insulator defects. The application is applied to the technical field of insulator defect detection.

The application discloses a 3D target detection method, a computer program product and a terminal, and belongs to the technical field of image processing. A laser radar and an image acquisition device are jointly calibrated. A point cloud global feature is mapped to an image feature map using a transformation kernel, a sparse matrix is constructed, and the image feature map is flattened into a matrix. After batch normalization processing of the sparse matrix and the matrix, connection processing is performed to obtain fusion features of the image and the point cloud. The fusion features are used for region proposal to generate a candidate target region. The candidate region is classified and a bounding box is regressed to determine the category and position of the target, and a 3D target detection result is obtained. The mapping relationship between the point cloud and the monocular image is constructed by using the sparse point cloud, so that only image pixel point coordinates and point cloud coordinate pairs that share the same points are paired, accurate alignment of the features is realized, the accuracy of target detection is improved, and the precision of 3D target detection is ensured.

This invention discloses a method and system for stacked forging instance segmentation based on an improved Mask R-CNN. The method includes: inputting the image of the stacked forging to be detected into a trained DSFE-Mask network; first, inputting the forging image into an MResNet feature extraction network to generate multi-level feature maps; inputting the multi-level feature maps into an AFPN feature progressive fusion network to output multi-scale feature maps; inputting the multi-scale feature maps into a Region Proposal Network (RPN) to obtain candidate regions; mapping the candidate regions onto the feature maps of corresponding scales through a Region of Interest (RoI) Align layer; generating a high-precision instance segmentation mask on one hand, and outputting target category information and corresponding detection boxes on the other hand; and outputting the instance segmentation result of the stacked forging. This invention significantly improves segmentation accuracy and robustness, and experimental verification on an industrial stacked forging dataset shows a significant improvement in mAP.

The application relates to a small farmland image segmentation method and device based on a double attention mechanism. The method comprises the following steps: acquiring a farmland image to be segmented; uniformly cutting the size of a collected farmland image to be segmented; inputting the farmland image to be segmented with a uniform size into a reconstructed Mask R-CNN model to obtain an instance segmentation result of the farmland image; wherein the main structure and the double attention mechanism feature pyramid in the reconstructed Mask R-CNN model are used for image feature extraction, a region proposal network is used to generate a plurality of regions of interest, the regions of interest are mapped to generate a fixed-size feature map through RoIAlign, and the three branches of the head of the Mask R-CNN model are used for prediction to obtain an object category, a refined bounding box positioning and an instance segmentation result. The method can realize accurate identification and instance segmentation of small farmland in a remote sensing image.

The invention discloses a remote sensing image small sample target detection model and method based on incremental learning, and the model takes a Faster R-CNN as a basic architecture, and combines a feature pyramid network (FPN) and a hybrid prototype comparison (HPC) coding module. Wherein the Faster R-CNN and the FPN are responsible for providing a multi-scale basic target detection capability; and the HPC coding module is used for uniformly coding the prototype vector of the basic category and the regional proposal characteristics of the new category so as to carry out effective comparative learning. The target detection method based on the model comprises the following two stages: (a) a basic training stage: carrying out comprehensive training on the model by utilizing sufficient basic category samples, and extracting prototype representation of each basic category; and (b) an incremental learning stage: on the premise of keeping part of network parameters frozen, performing fine adjustment on the model by using a small number of new category samples, and in combination with a hybrid prototype comparison coding module and a prototype calibration module, relieving the negative influence of disastrous forgetting and improving the detection performance of the new category.

The invention discloses a skylight mounting plate welding quality detection method based on a convolutional neural network, and the method comprises the following steps: obtaining image data of a detected part through an industrial camera, and inputting the image data into a model based on the convolutional neural network; the convolutional neural network is combined with a Sobe l edge detection operator, local texture and edge features of the welding surface are extracted, and the similarity between the features is calculated based on cosine similarity and used for recognizing fine texture changes of the welding surface; and extracting the shape of the welding spot and the geometric structure characteristics of the rivet nut through a region proposal network for identifying the dimensional deviation. A template matching method based on small sample learning is adopted. The model firstly extracts key features in input data, maps the key features to a unified feature space, and then performs similarity comparison by means of a pre-constructed template library to realize rapid defect type identification.

The application discloses an automatic driving perception method based on three-dimensional point cloud data, which comprises the following steps: S1, acquiring three-dimensional point cloud data; S2, processing the three-dimensional point cloud data and performing three-dimensional target detection by using an improved PV-RCNN++ network, which comprises the following steps: S21, voxelizing the point cloud data; S22, processing the voxel data by using a deformable sparse convolution; S23, stacking and aggregating multi-scale voxel features to generate a feature map, and using a region proposal network to generate a candidate region; S24, performing farthest point sampling on the candidate region; S25, encoding multi-scale voxel features by using a voxel set abstraction module and a spatial attention mechanism; S26, aggregating multi-scale voxel features, point cloud bird's eye view features and key point features by using a region of interest pooling module; S27, performing fine processing on the candidate region; S3, calculating the distance and speed of a target by using a deep learning network; S4, predicting a target motion trajectory by using Kalman filtering; and S5, outputting the target motion trajectory, so that the detection and prediction precision of various dynamic targets in automatic driving are improved.

The invention relates to the technical field of detection data processing, in particular to a textile fiber component analysis system and method based on machine learning. Before the direction gradient attention module provided by the embodiment generates the candidate region proposal, the feature representation has strong direction deviation, so that the region proposal network can be positioned to the geometric center more easily according to the extension direction of the fiber, thereby generating a more compact and more accurate candidate box, and improving the accuracy of the region proposal network. And a foundation is laid for subsequent adhesion fiber unwrapping. Attention is focused on several main directions related to fibers, the directional gradient attention module naturally suppresses edge response from random direction background noise, features are purified, and the robustness of the model is improved.

This invention belongs to the field of target classification technology, and particularly relates to a method and system for analyzing 3D point cloud data based on artificial intelligence. It includes steps such as data acquisition, data preprocessing, model building, model training, and model prediction. By acquiring raw point clouds and performing intelligent denoising and completion operations, the integrity and accuracy of the point cloud data are effectively improved. A deep model is constructed using an input layer, a region proposal layer, and classification and regression layers to accurately classify targets and predict 3D bounding boxes. The region proposal layer combines seed point feature extraction and a voting mechanism to generate candidate regions; the classification branch outputs class probabilities, and the regression branch predicts bounding box parameters. This modular system has advantages such as strong noise suppression, high completion accuracy, high target detection accuracy, and applicability to multiple scenarios, making it particularly suitable for the efficient automatic recognition of complex 3D structures.

This invention provides a robot dynamic grasping method and system based on multimodal data fusion, comprising: performing multimodal fusion of visual and tactile data of a dynamic object to be grasped by the robot to obtain multimodal fused data; based on the multimodal fused data, performing target recognition on the dynamic object to be grasped using a region proposal-based target detection algorithm to obtain position and pose information; based on the position and pose information, performing state prediction on the dynamic object to be grasped to obtain state prediction information; and based on the state prediction information, generating a grasping path for the robot using a reinforcement learning algorithm. This application, by performing multimodal fusion of the collected visual and tactile data, enables the robot to obtain comprehensive environmental information, increasing the robot's adaptability to environmental changes; and by predicting the state of the object to be grasped, the robot can predict the object's trajectory, which helps improve the robot's grasping success rate.

A target detection system, a target detection method, a corresponding training method, an electronic device, and a readable storage medium, the target detection system can include: a backbone network, a neck network, and a head network, the backbone network is configured to extract different scale features of an input image, and generate a multi-layer feature map; the neck network is configured to perform feature fusion on the multi-layer feature map to generate a multi-layer feature fusion map; the head network includes an adaptive region proposal network and a detection network, wherein the adaptive region proposal network is configured to obtain initial region proposals about a detection target based on each layer of the multi-layer feature fusion map, and adaptively adjust each initial region proposal to obtain a final region proposal about the detection target in each layer of the multi-layer feature fusion map; the detection network is configured to perform position regression and classification on the detection target based on the final region proposal to obtain the position and category of the detection target in the input image.