18results about How to "Improve object detection accuracy" patented technology

The invention discloses a visual perception and trajectory restoration method, system and device for chicken motion analysis and anomaly detection in a large cage farm and a storage medium, and relates to the technical field of video monitoring and behavior analysis. Carrying out fine feature extraction by adopting a space and channel attention mechanism; fine features are combined with Kalman filtering and a Hungary algorithm, chicken individual identities are continuously tracked and adjusted through time sequence prediction and space matching, and a target bounding box is output; adjusting the individual identity of the chicken according to tracking, performing smooth repair on interrupted and shielded trajectory segments through spline interpolation and Kalman filtering, and performing reasonable trajectory recovery by adopting biological kinematics constraint; based on the target bounding box, a frequency domain feature analysis mechanism and a deep learning coding technology are introduced, a behavior feature vector is generated, and health early warning is triggered according to the abnormal confidence coefficient; according to the method, the chicken monitoring precision and the anomaly detection reliability are improved.

The application provides a multi-modal target detection method based on super-resolution assistance, and the implementation steps are as follows: obtaining a training sample set and a test sample set; building a multi-modal target detection network model based on super-resolution assistance; iteratively training the multi-modal target detection network model based on super-resolution assistance; and obtaining a target detection result. The multi-modal fusion module is provided with a symmetrical double-branch structure, realizes efficient fusion of RGB and IR images at the input end, retains color integrity and complementary information of thermal radiation, significantly improves feature discrimination, and effectively improves detection accuracy; the super-resolution assistance branch serves as an internal branch, directly optimizes feature representation of the backbone network through a reconstruction loss, thereby significantly enhancing the expression capability of the network for small targets and fine features, and meanwhile, the super-resolution assistance branch does not participate in the detection process of the test sample, realizes zero additional calculation overhead in the reasoning stage, and improves the detection efficiency while improving the target detection accuracy.

The application discloses a kind of training and target detection method and device of target detection model, belong to the field of automatic driving, the method includes: according to the sample multi-view image in the sample scene where the image acquisition equipment is collected in the sample scene where the automatic driving vehicle is located, determine sample image dense feature;According to the sample point cloud data in the sample scene where the automatic driving vehicle is located that remote sensing detection equipment is collected, determine sample image sparse feature;Sample image dense feature and sample image sparse feature are fused, and sample fusion image feature is obtained;According to sample fusion image feature, target detection model is trained.The application solves the problem that feature is not aligned, enhances the feature fusion effect, and then improves the accuracy of subsequent target detection result.

The application discloses an underwater target detection encoder, a detection system and a method, the encoder comprising an input projection layer, a global semantic modeling unit, a first-stage feature fusion unit and a second-stage feature fusion unit; wherein the input projection layer is used for channel mapping of multi-scale features output by a backbone network, the global semantic modeling unit establishes a dependency relationship between spatial positions in a feature map through a self-attention mechanism; the first-stage feature fusion unit introduces a frequency domain transformation path in a cross-scale feature fusion process, performs frequency screening on features and generates high-frequency perception features; the second-stage feature fusion unit introduces intermediate features in the first stage in a propagation process and performs splicing and reconstruction processing, forming multi-scale feature representation. Based on the encoder, the detection system and the method are constructed, realizing classification and positioning processing of targets in underwater images.

This application discloses a target detection method, apparatus, device, and storage medium. The method includes: acquiring and processing a video to be detected containing a dynamic background to construct a differential image dataset; dividing the differential image dataset to obtain a target image dataset and a background image dataset; passing the target image dataset through a trained first detection model to obtain a first detection result, and passing the background image dataset through a trained second detection model to obtain a second detection result; filtering and performing frame interpolation on the second detection result to obtain a second target detection result; and obtaining the detection result of the target relative to the background information based on the first and second target detection results. This scheme can filter out false detections and supplement lost frame results after detecting different types of targets, so as to accurately determine the detection result of the target relative to the background information, thereby improving the accuracy of target detection.

The invention discloses a drop-out fuse state monitoring method based on multi-size features and deep learning, and the method comprises the steps: S1, forming a data set for obtained drop-out fuse pictures, and dividing the data set into a training set and a test set; s2, adding an improved receptive field block and a coordinate attention module to a YOLOx backbone network, adding an adaptive spatial feature fusion module to PANet, carrying out secondary fusion on features of different scales, then introducing a loss function of weighting loss and positioning loss fusion, and finally carrying out lightweight improvement, constructing a state monitoring model, and carrying out state monitoring. Performing training verification on the constructed state monitoring model through the training set and the test set; s3, identifying a drop-out fuse picture acquired in real time by adopting the trained and verified state monitoring model; according to the image data collected by the application, the construction of a special database is realized, the YOLOx is improved and lightweight operation is carried out, and the complexity of the model is reduced, so that the method is suitable for an embedded platform of an electric unmanned aerial vehicle, and the detection effect of the drop-out fuse is improved.

This invention provides an automatic SAR target detection method and apparatus based on adaptive feature focusing, belonging to the field of SAR image detection. The method includes: constructing a lightweight EOD-Net; after the backbone network extracts multi-scale features, the latent target perception branch PTBranch outputs a mask to coarsely locate the target region; and using a no-overlap feature map cropping NOFC to iteratively remove overlapping blocks, retaining only non-overlapping target feature blocks for the detection head to regress. During the training phase, a Gaussian-binary hybrid mask supervises PTBranch, and the focus loss and DICE loss are jointly optimized. This invention outperforms comparable methods such as YOLOv11 and Faster RCNN, achieving high-precision and efficient SAR image detection.

The invention provides a back splint wearing condition detection method and device, equipment and a storage medium. According to the detection method, an image detection target based on a mine scene is a small-size target object related to a back splint (for example, narrow and thin straps when a person wears the back splint on the front side, shielding is caused by changeable postures of the person, and only a small part of area of the back splint can be exposed in an image); an image enhancement mechanism which is beneficial to highlighting the small-size target object in the to-be-detected image is designed in a targeted manner, so that target detection is carried out on the target enhanced image after image enhancement, and the image target detection accuracy of the back splint can be effectively improved.

This application provides a robust visual 3D object detection method for multiple scenarios based on lighthouse cognition, relating to the field of autonomous driving technology. The method includes: acquiring an RGB image of the target scene; processing the RGB image using a pre-trained backbone network to obtain a first image feature map; processing the first image feature map using a pre-trained dynamic deep network to obtain a semantic feature map and a depth distribution map; multiplying the semantic feature map and the depth distribution map to obtain a second image feature map; processing the second image feature map using voxel pooling to obtain a BEV feature map; and processing the BEV feature map using a detection head to obtain a 3D object detection result. This application improves the accuracy of object detection in multiple scenarios.

The application provides a roadside target detection method and device for travel map danger warning, the method comprises the following steps: processing a roadside image by using a feature extraction model to obtain visual features and ground plane features; processing the visual features by using a visual encoder to obtain visual embedding features; processing the ground plane features by using a ground plane encoder to obtain ground plane embedding features; processing the visual embedding features and the ground plane embedding features by using a ground plane guided decoder to obtain a ground plane perception object query; processing the ground plane perception object query by using a detection head to obtain a target detection result; processing the ground plane features by using a ground plane predictor to obtain a ground plane equation graph; and calculating a target depth value by using two-dimensional pixel coordinates of a target bottom surface center, the ground plane equation graph and camera parameters. The application effectively improves the target detection accuracy of the roadside monocular image and provides a key data basis for the danger warning system of the travel map.

The invention discloses a millimeter wave radar and vision fused real-time sensing system and method. The method comprises the following steps: step 1, acquiring environment original data through a millimeter-wave radar and a visual sensor; 2, preprocessing the radar data obtained in the step 1 to generate point cloud; step 3, preprocessing the visual data obtained in the step 1 and performing feature extraction to obtain multi-scale visual features; 4, performing time synchronization and space alignment on the point cloud and the multi-scale visual features; 5, BEV spatial feature coding is carried out on the aligned radar point cloud and visual features; step 6, performing cross-modal feature fusion on the coded radar point cloud and visual features; and step 7, based on the fused features, realizing 3D target detection and positioning. In combination with signal processing and deep learning technologies, efficient sensing is realized through data preprocessing, synchronous alignment, fusion algorithm design and real-time optimization.

The application relates to a weakly supervised target detection method based on an adversarial collaborative learning, which comprises the following steps: acquiring image data to be detected, inputting the image data into a pre-trained adversarial collaborative network, and outputting a target detection result, wherein the adversarial collaborative network comprises two peer network models with the same structure, each peer network model is a standard WSOD model, the WSOD model comprises a candidate region feature extractor and a task head, and the task head comprises a multi-instance detection module, an online instance classification module and a detection head. Compared with the prior art, the application has the advantages of reducing the dependence on a large amount of accurate labeled data, improving the perception ability of complete objects and the like.

The invention provides a lightweight multi-scale detection method and system for unmanned aerial vehicle small target aerial photography, and the method comprises the steps: obtaining visible light or infrared image data of unmanned aerial vehicle aerial photography, carrying out the preprocessing of an image, and obtaining a preprocessed image; constructing a detection model based on a YOLOv11 model, extracting multi-scale features, and performing weighted fusion processing on the multi-scale features to obtain a cross-layer information aggregation feature map; constructing an unmanned aerial vehicle aerial photography small target data set, obtaining a training set sample, inputting a detection model, optimizing model parameters, and obtaining a lightweight multi-scale detection model; performing parallel detection on the cross-layer information aggregation feature map based on a lightweight multi-scale detection model, and outputting a detection result; the detection result is transmitted to the terminal in real time according to a set mode; the feature extraction capability is enhanced by constructing a detection model, effective complementation of shallow spatial information and deep semantic information is realized, a multi-scale feature map with rich semantic information is generated, and the small target detection precision is improved.

The application is suitable for the technical field of target detection, and provides a fusion method and device of millimeter wave radar data and images in a high-speed scene, in which two millimeter wave radars arranged back to back and a variable-focus spherical camera are combined and arranged to realize target detection in a highway scene. First, two millimeter wave radars can obtain radar data of two-way highways, visual perception data of each focal length can be obtained by adjusting the focal length of the spherical camera, and finally the perception data of the millimeter wave radar and the spherical camera are fused to obtain the final target detection result. Thus, the variable-focus spherical camera ensures that there is no visual blind area in the middle and long distances, so as to control the visual perception range to be consistent with the radar perception range, and improve the target detection precision.

The application discloses a kind of offshore small target detection methods based on multi-domain multi-feature under high sea condition.Firstly, the characteristics of sea clutter and target signal in time domain and frequency domain are analyzed using the relevant information of existing sea clutter data.Under the condition of low signal-to-clutter ratio of sea clutter data, three key features are extracted combining with fractal knowledge, which are fractional order Riemann-Liouville (RL) integral domain H index, instantaneous phase entropy and fractional Fourier transform (Fractional Fourier Transform, FRFT) domain relative peak height, wherein the H index is calculated using Detrended Fluctuation Analysis (DFA) method.Finally, the small target detection is realized using the bidirectional time series modeling advantage of Bi-GRU (Bidirectional Gated Recurrent Unit) model, and the effectiveness of the model is verified through multi-dimensional performance indicators.The application makes full use of the physical properties of sea clutter, combines fractional integral, entropy and fractal method, realizes sea surface target detection under the condition of low signal-to-clutter ratio, and improves the target detection accuracy;At the same time, the Bi-GRU model is efficient and has strong information capturing ability, can handle complex and variable sea clutter environment, has wide applicability and practical value.

The invention discloses a remote sensing image small target detection method based on scale and frequency adaptive perception, and relates to the technical field of remote sensing image small target detection. The method comprises the following steps: firstly, organizing an image file and a corresponding annotation file to obtain image data, and dividing the image data into a training image and a test image to form a required data set; a training set image is subjected to standardization preprocessing and then input into a network, multi-scale features are extracted through a frequency self-adaption fusion module, an anchor frame matched with a target size is generated in combination with a scale self-adaption anchor frame generator, then a loss function is calculated, model parameters are optimized through end-to-end training, and finally the optimal weight is stored. And finally, reasoning the images in the test set by using the model weight obtained by training, and checking the small target detection capability of the model in an actual scene. According to the method, the label distribution mechanism is jointly optimized, the feature representation capability of the small target is enhanced, and the remote sensing image small target capability is effectively improved.

The invention belongs to the related technical field of PCBA (Printed Circuit Board Assembly) quality detection, and discloses a construction method, a fine adjustment method and a system of a task decoupling small sample target detection model, comprising the following steps of: (1) extracting frozen single-scale features of an image by adopting a frozen DINOv2 backbone network; respectively carrying out projection and dimension reduction on the frozen single-scale features by adopting two neck networks with the same structure and non-shared parameters so as to obtain to-be-decoded features and prompt features; (2) a class-independent increment positioning prompter is adopted to search targets possibly existing on the image according to the prompt features, a target prompt box and a confidence score are formed, and then query features are screened out; and (3) a task heterogeneous decoupling decoder is adopted to predict a prediction label corresponding to each query according to the to-be-decoded features, the target prompt box and the query features, the mixed loss of the prediction labels and true value labels is calculated, and then non-freezing parameters of the small sample target detection model are optimized. According to the invention, the detection precision of the model in the small sample scene is improved.

This application discloses a target detection method, apparatus, device, and readable storage medium. The method includes: acquiring a current frame image; performing density detection on targets in the current frame image to obtain density detection results; cropping the current frame image according to the density detection results to obtain image blocks; inputting each image block into a target detector corresponding to the target density in the image block for target detection; and stitching together the target detection results of each image block to obtain the target detection result of the current frame image. The above-disclosed technical solution dynamically and adaptively divides the current frame image into blocks based on the target density distribution, and selects a target detector adapted to the image block based on the target density distribution in different image blocks for target detection, thereby improving target detection capability and accuracy. Furthermore, density detection of the current frame image can remove non-target regions, thus improving target detection speed.