11 results about "Pyramid algorithm" patented technology

This invention relates to a multi-agent collaborative decision-making method and apparatus, belonging to the field of multi-agent collaborative decision-making technology. The method includes: each agent jointly sensing and encoding the acquired multimodal input data through a data perceptron; using a data updater employing a time-series incremental consistent ensemble algorithm to fuse and update the current multimodal representation vector with the historical representation set; compressing the fused encoding through a data transmitter and sending it to neighboring agents; each agent receiving the encoding information from neighboring agents through a data receiver and aligning and fusing it with its own fused encoding; and the data updater updating the fused representation using a multi-level weighted feature pyramid algorithm to generate new historical representation data to support the next round of collaborative decision-making. This invention improves the reliability and robustness of multi-agent collaborative decision-making through four core components: a data perceptron, a data updater, a data transmitter, and a data receiver.

The invention belongs to the technical field of computer vision and industrial safety monitoring, and particularly relates to a large hangar door operation area foreign matter detection method based on multi-scale feature fusion, and the method comprises the following steps: S1, collecting a real-time gray level image of the ground of a hangar door operation area, carrying out the preprocessing of the real-time gray level image, and obtaining a real-time gray level image; carrying out multi-scale decomposition on the preprocessed image by utilizing a Laplacian pyramid algorithm, and constructing a feature pyramid containing multiple layers of feature maps; s2, establishing a mapping relation between row coordinates of the preprocessed image and a physical scale by using installation geometric parameters of the camera, and calculating perspective scale scaling factors at different row coordinates of the preprocessed image; by reducing the weight of the high-frequency layer and utilizing the low-frequency structure layer, high-frequency texture noise such as ground oil stains and scratches is naturally filtered, the false alarm rate is effectively reduced, the calculation complexity is low, the method can be directly deployed on a low-power-consumption industrial personal computer, and the engineering implementation cost is reduced.

The invention relates to a multi-agent collaborative decision-making method and device, and belongs to the technical field of multi-agent collaborative decision-making. The method comprises the following steps: each agent performs joint perception and feature coding on acquired multi-modal input data through a data perceptron; fusing and updating the current multi-modal representation vector and the historical representation set by using a data updater and adopting a time sequence increment consistent integration algorithm; the fused codes are compressed through a data transmitter and sent to an adjacent agent; each agent receives coding information from an adjacent agent through a data receiver, and the coding information is aligned and fused with a fusion code of the agent; and the data updater adopts a multi-level weighted feature pyramid algorithm to update the fusion representation, and new historical representation data is generated to support the next round of collaborative decision. According to the multi-agent collaborative decision-making method, the reliability and robustness of multi-agent collaborative decision-making are improved through the four core components including the data sensor, the data updater, the data transmitter and the data receiver.

The application discloses a kind of civil aviation weather short-time prediction methods based on weather radar echo extrapolation, by combining the optical flow method of pyramid algorithm to construct the motion vector field of radar echo map of radar wave, on the basis of motion vector field, extrapolation is obtained to predict the radar echo map of future time, and fusion ADS-B terminal system is displayed and early warning.The method realizes 2 hours extrapolation prediction of weather radar echo, helps weather, control and flight personnel to grasp the reference of weather;Realize a weather, control and flight organic integration of diversified platform, conducive to weather, control, flight personnel to establish consistent scene consciousness, guarantee flight safety.Control and flight personnel can effectively, timely obtain the latest meteorological information and dangerous weather warning information, and combine with ADS-B technology, according to aircraft track and meteorological warning information, guide aircraft to avoid dangerous weather, so as to improve airspace utilization.

The invention relates to a digital ray image multi-energy spectrum fusion component defect detection method, which comprises the following steps of: acquiring a plurality of original digital images of a workpiece under different pipeline voltages by adopting a digital ray imaging detection system, and performing image fusion by taking the original digital images as source images and adopting a weighted least square filter algorithm and a pyramid algorithm, and obtaining a fused defect detection result graph. According to the method, a weight is distributed to each image observation data point through a weighted least square filter method, a smooth and optimized image is obtained, feature information of a source image is extracted from the obtained smooth image from different scales and different spatial resolutions in cooperation with a pyramid algorithm, and then the extracted feature information is fused to obtain a fused image; therefore, the images of the excessive exposure place and the insufficient exposure place are clearly displayed, so that the defects of the component are efficiently and quickly detected.

The application provides a mold monitoring method and system based on deep learning and multi-scale template matching, and a medium, the method comprising the following steps: constructing a multi-angle multi-size template, acquiring a real-time collected template image or mold image, and pre-processing and enhancing the collected image; performing image matching on the pre-processed and enhanced image through a pyramid algorithm; judging whether the template or mold has defects according to the matching result; if so, performing morphological processing and Canny edge extraction on the defect area, and performing false detection filtering based on geometric features to remove abnormal areas; and positioning the defect area of the template or mold through connected component analysis. Through the fusion of multi-template dynamic management, multi-scale pyramid matching, image enhancement preprocessing, refined post-processing and semi-automatic linkage mechanism, the application realizes millisecond-level response and stable detection in a high-speed and complex environment.

The invention relates to the field of image processing, and discloses a general camera near-infrared image generation method, which comprises the following steps: S1, adaptive sub-pixel image registration: acquiring two frames of images, namely an image under an open filter and an image under a cut-off filter, the two frames of images are subjected to sub-pixel-level alignment based on an optical flow method or a feature point pyramid algorithm, in the method, the sub-pixel alignment technology based on the optical flow method or the feature point pyramid is adopted, the precision reaches 0.1 pixel level, and nonlinear deformation caused by turbulence can be eliminated; a compensation coefficient is estimated online in real time through a spectral decoupling and dynamic gain compensation model, and the problems of exposure difference and non-uniform spectral response are solved; edge enhancement and noise suppression are performed by combining bilateral filtering or guided filtering with a visible light image as guidance, clear structure information is transferred to a near-infrared image, and meanwhile, picture jitter is suppressed by using low sensitivity of a near-infrared band to turbulence.

A crater identification method and device for landing visual navigation, comprising the following steps: establishing a crater pair database: selecting a proper number of craters to establish a database, removing small craters and overlapping craters according to selection principles to obtain a crater pair table; screening and matching the crater pair database; and re-projecting and verifying the screened and matched crater pairs and estimating the current pose. The database of the present application only stores the serial numbers of the crater pairs and corresponding invariants, greatly reducing the database capacity and improving the search efficiency. The present application proposes a dynamic threshold to solve the problem that the projective invariant is sensitive to noise, improving the matching rate and average processing speed. Meanwhile, the iterative pyramid algorithm is proposed to quickly screen the redundancy, estimate the pose on the basis of preliminary identification, and verify the identification result after re-projecting and matching. The present application has the characteristics of high identification rate, robustness to noise, and accurate and fast pose estimation, reaching the leading performance level of initial identification in landing navigation.

The invention belongs to the technical field of image processing, and provides a pipeline construction risk identification method based on an unmanned aerial vehicle and a YOLOv5 algorithm, and the method comprises the steps: collecting an original image; analyzing sampling weights of different areas of the original image; according to the sampling weight, performing different-scale sampling on the original image through an image pyramid algorithm to obtain a sampling image; segmenting and analyzing the sampled image through a YOLOv5 algorithm to obtain significance features of a plurality of target grid blocks; clustering the target grid blocks according to the saliency features to obtain a second clustering region of the original image; and analyzing saliency characteristics of the second clustering region to obtain a saliency region, and performing pipeline construction risk identification according to the saliency region. The original image is sampled through the image pyramid sampling algorithm, the sampling weights of different areas are determined according to changes of different details in the original image, detail information in the original image is highlighted, and the small target detection accuracy of the YOLOv5 algorithm is improved.

This disclosure provides a multi-view stereo matching method based on optical flow consistency, belonging to the field of image processing technology. Specifically, it includes: performing downsampling using an image pyramid algorithm; calculating the depth map and normal vector map of each downsampled image; selecting the best neighboring image; calculating the optical flow motion vector of the downsampled image based on the dense optical flow algorithm and the best neighboring image; constructing optical flow consistency; jointly estimating the optical flow consistency and photometric consistency; recalculating the depth map and normal vector map of each downsampled image according to a cost function; comparing the obtained depth map and normal vector map using a detail restorer, and selecting the depth value and normal vector for each pixel using a selection function to obtain the final depth map and normal vector map for the next layer; finally, fusing the depth map and normal vector map into a three-dimensional point cloud model. This disclosure improves the accuracy and adaptability of image matching.

The present application relates to a kind of fusion personnel behavior's college laboratory computer room safety early warning method, belong to computer room safety early warning technical field.It includes: by collecting the original video stream in target university laboratory computer room;Original video stream is framed and handled, and feature recognition is carried out using pyramid algorithm to obtain the feature image of each video frame;The video frame corresponding to the feature image containing personnel feature is as key video frame, and the personnel identifier and its personnel joint in each key video frame are determined;According to all personnel joints, the time sequence motion characteristics corresponding to each personnel identifier are generated;According to the time sequence motion characteristics corresponding to each personnel identifier and behavior area, the personnel behavior corresponding to each personnel identifier is determined;Generation fusion personnel behavior's target university laboratory computer room safety early warning information.The present application can reduce the personnel misoperation, equipment damage and data loss etc. that target university laboratory computer room can appear possibly.