12results about How to "Improve target recognition accuracy" patented technology

The invention relates to the technical field of unmanned aerial vehicle control, and discloses an unmanned aerial vehicle target classification matching strike control method and system based on deep learning, and the method comprises the following steps: S1, collecting the original environment data of an unmanned aerial vehicle flight region, carrying out the preprocessing of the original environment data, and generating multi-mode perception data; and S2, inputting the multi-modal sensing data into a pre-constructed deep learning classification network, extracting multi-level depth features of the target through the deep learning classification network, carrying out classification identification on the target based on the multi-level depth features, and outputting target category information and target position information of the target. The multi-modal sensing data is input into the deep learning classification network for target recognition, the multi-modal fusion sensing mode can give full play to the complementary advantages of different sensors, high target recognition accuracy can still be kept in complex environments such as night, low illumination and severe weather, and the adaptive capacity of the system to environment changes is remarkably improved.

The present application relates to the technical field of image processing, in particular to a fisheye lens image pixel-level edge enhancement and denoising method, comprising: a data acquisition step: acquiring original fisheye image data and preset fisheye lens optical distortion parameters; a weight generation step: determining the distortion stretching rate of a pixel point according to the optical distortion parameters; and generating a spatial density weight map; a threshold determination and denoising step: determining a target denoising threshold according to the spatial density weight map; denoising the original fisheye image data to generate an intermediate denoising image; a convolution kernel generation step: generating an adaptive curved surface convolution kernel according to the optical distortion parameters; an edge enhancement step: using the adaptive curved surface convolution kernel to perform edge enhancement processing on the intermediate denoising image; and generating an enhanced image; the present application effectively avoids the risk of irreversible erasure of key semantic features in the edge field, and ensures high-reliability feature preservation in the extreme field.

The application discloses a kind of soft membrane circuit board defect detection method and related device, the method includes: obtaining defect sample data set, the defect sample data set includes the circuit board sample image of multiple different defects;YOLOv7 target detection algorithm is used, part of data in the defect sample data set is used as training set, another part of data in the defect sample data set is used as verification set, and the training and verification of detection model are carried out;Get the detected circuit board image, and the template matching mode is used to accurately position each circuit part in the detected circuit board image, then the detection model is used for defect detection.According to the soft membrane circuit board defect detection method and device of the application, YOLOv7 target detection algorithm is used as basic detection model, is not sensitive to the change caused by transmission machine table shaking or unstable light source during image shooting, does not need to set a large number of processing parameters, and has higher target recognition accuracy.

The invention discloses a bird and unmanned aerial vehicle identification method based on compensated multi-frame target slice and flight path fusion, and the method specifically comprises the steps: firstly, obtaining a multi-frame R-D domain slice data set of a rotor unmanned aerial vehicle and a bird through the staring distance compensation and Doppler compensation operation; secondly, a bidirectional GRU network model fusing track information is adopted, multi-frame target R-D domain slice data are input, and target feature extraction and integration are completed through double-layer convolution and a full connection layer; then, target track features are added into the integrated target feature set, and the target track features and the integrated target feature set are input into a bidirectional GRU network together; and finally, completing the output of a target identification result label through a full connection layer, and completing an identification task. According to the method, the track information and the multi-frame target information are combined, the track features and the motion features of the target are fully utilized, the target recognition accuracy is improved, and the method is simple, high in recognition speed and high in real-time performance.

The present application relates to the field of image processing and target image recognition technology of civil optical equipment such as intelligent traffic, city supervision and commercial monitoring, and particularly relates to a multi-target characteristic recognition method based on photoelectric image, which aims to improve the adaptability and accuracy of target recognition in target optical imaging, and the method obtains target motion image through photoelectric equipment, divides the target into relative static, straight line motion and turning state according to a preset threshold, extracts target motion trajectory, judges target state, respectively carries out straight line and arc line fitting on the target trajectory of straight line motion and turning state, analyzes the overall motion state of multiple targets in the imaging image based on the motion state trajectory, and adopts corresponding individual motion trajectory statistical analysis method for different states, the multi-state adaptive analysis capability of the method significantly improves the recognition accuracy, and the recognition accuracy of the target with frequently changing state in a complex scene is improved by more than 30%.

The present application relates to a kind of multi-modal visual large model under the dynamic target identification system of heterogeneous sensor anti-interference.The system includes: anti-background interference module, by modality convolution to multi-modal data of heterogeneous sensor, LSTM time series modeling and weighted fusion generation space-time feature, combine time / space attention mechanism and Transformer reinforced feature, use Gaussian probability density model to filter foreground pixel and extract foreground feature;Multi-modal processing module, separate each modality foreground feature and splice after refining, by Transformer extract consistent feature and complementary feature, after fusion and pooling enhancement generation reinforced fusion feature;Target identification module, fusion foreground feature and reinforced fusion feature, after convolution compression and Timesformer long time series modeling, input target detection head, output target classification and boundary box.The present application solves the problem of heterogeneous sensor data interference suppression and dynamic target accurate identification, improves robustness in complex scene.

The invention relates to a road target real-time detection and early warning system in a foggy environment. The system comprises a detection parameter determination module which is used for obtaining visibility parameters of a foggy environment and determining target detection parameters of sensing equipment based on the visibility parameters; the sensing equipment configuration module is used for configuring sensing equipment by adopting the target detection parameters and executing road target detection based on the configured sensing equipment to obtain a first target detection result; the target detection optimization module is used for performing detection optimization through a machine learning model based on the first target detection result to obtain a second target recognition result; the target attribute calculation module is used for calculating a target bounding box and a target motion trend based on the second target recognition result; and the risk early warning generation module is used for judging a target risk level according to the target bounding box and the target motion trend and generating a risk early warning signal. By adopting the system, the accuracy and the real-time performance of road target detection in a foggy environment can be improved.

The invention relates to the technical field of image processing, in particular to a fisheye lens image pixel-level edge enhancement and denoising method, which comprises the following steps: a data acquisition step: acquiring original fisheye image data and preset fisheye lens optical distortion parameters; a weight generation step: determining the distortion stretch rate of the pixel points according to the optical distortion parameters; generating a spatial density weight map; a threshold determination and denoising step: determining a target denoising threshold according to the spatial density weight map; de-noising the original fisheye image data to generate an intermediate de-noised image; a convolution kernel generation step: generating a self-adaptive curved surface convolution kernel according to the optical distortion parameters; an edge enhancement step of performing edge enhancement processing on the intermediate de-noised image by using an adaptive curved surface convolution kernel; generating an enhanced image; the risk that key semantic features of the edge view field are irreversibly erased is effectively avoided, and it is ensured that high-reliability feature retention is achieved in the extreme view field.

This invention relates to a method, apparatus, and computer-readable storage medium for underwater acoustic multi-target identification, belonging to the field of underwater target identification technology. The invention includes: calculating the detection features of an underwater acoustic audio signal, and inputting the detection feature vector into a first classifier for classification to determine whether a corresponding underwater acoustic target exists in the audio signal; if so, calculating the identification features of the audio signal and inputting the identification feature vector into a second classifier to determine the type of underwater acoustic target corresponding to the audio signal. This invention can directly perform feature calculation on the underwater acoustic audio signal, offering high real-time performance and fast response; and by analyzing the underwater acoustic audio signal from multiple angles to obtain detection and identification features, it reduces information loss during feature extraction; furthermore, it divides underwater acoustic target identification into two stages: detection and identification. First, the underwater acoustic target is detected based on the detection features, and then the underwater acoustic target type is identified based on the identification features, improving the accuracy of underwater acoustic target identification.

The invention discloses an intelligent bag taking method and system based on target recognition analysis, and relates to the technical field of target recognition. The method comprises the following steps of bag taking interference analysis, bag taking identification analysis and bag taking decision feedback. According to the method, the bag taking interference analysis result is obtained based on the bag taking identification interference parameters so as to judge whether to carry out image dynamic acquisition adjustment or not, if yes, bag taking dynamic identification analysis is carried out after the image dynamic acquisition adjustment is carried out, and if not, bag taking dynamic identification analysis is directly carried out; recognition errors caused by ambient light changes or other factors are effectively avoided, recognition of the bag taking action is more accurate, whether a bag taking decision feedback mechanism is triggered or not is judged after bag taking dynamic recognition analysis is conducted, if yes, bag taking decision feedback is executed, and if not, image dynamic recognition optimization is executed, so that the bag taking action is more accurate. The reliability of the automatic bag taking machine in actual operation is effectively improved, and then recognition of the bag taking process of the automatic bag taking machine is more accurately carried out.

This application discloses a laser mapping method and apparatus, and a laser positioning method and apparatus. The laser mapping method includes: acquiring first image data and first laser point cloud data collected by a first vehicle in a preset operating area, and fusing the two data using a preset fusion algorithm to obtain a first fusion result; determining the target and its corresponding target type in the first laser point cloud data based on the first fusion result; preprocessing the target in the first laser point cloud data according to the target type using a preset preprocessing strategy to obtain preprocessed first laser point cloud data; and constructing a point cloud map of the preset operating area based on the preprocessed first laser point cloud data and corresponding post-processed positioning data. This application fuses image data and laser point cloud data for target recognition, improving target recognition accuracy. It also employs different preprocessing operations on the laser point cloud data for different target types, thus improving the accuracy of laser mapping and positioning.

Embodiments of the present application provide a target comprehensive identification method and device based on high-dimensional feature map, and equipment, the method comprises: obtaining observation data about target group based on radar sensor collection; processing the observation data, and performing feature extraction on the processing result to obtain multi-dimensional feature data about the target group; processing the multi-dimensional feature data to form a target multi-dimensional feature vector; inputting the target multi-dimensional feature vector into a self-encoder to realize data dimension reduction and deep feature extraction of the target multi-dimensional feature vector based on the self-encoder, and obtaining a target multi-dimensional feature map; inputting the target multi-dimensional feature map into a target neural network to classify the target multi-dimensional feature map based on the target neural network, and realizing target identification. The method based on the embodiments can effectively improve the accuracy and efficiency of target identification.