12 results about "Sample distance" patented technology

The invention discloses a multi-scene universal pedestrian re-identification method and system based on scene semantic guidance feature decoupling, and the method comprises the steps: carrying out the standardization processing and feature coding of an input pedestrian image, and generating a global feature containing pedestrian and scene information; through a pedestrian feature encoder and a scene feature encoder which are mutually independent, the global features are decoupled into pedestrian identity features and scene features; splicing the pedestrian identity features and the scene features and then inputting the spliced features into a gating network to generate an adaptive weight value; performing normalization processing on the adaptive weight value to obtain a scene feature weight and a pedestrian feature weight; performing weighted fusion on the decoupling features according to the scene feature weight and the pedestrian feature weight to generate classification features; on the basis of the classification features, a sample distance is calculated by adopting a self-adaptive marginal triple loss algorithm, and classification information is output; and based on the classification information, deploying a training model to a camera through knowledge distillation to realize pedestrian re-identification. According to the invention, the recognition precision in a complex scene is improved.

The invention provides a photovoltaic inspection method and device, electronic equipment and a storage medium, and relates to the technical field of unmanned aerial vehicle photovoltaic inspection, and the method comprises the steps: obtaining the geographic information, photovoltaic array parameters and inspection task parameters of a photovoltaic station; wherein the inspection task parameters comprise inspection time and an expected sampling distance; determining a mirror reflection area of the photovoltaic array according to the geographic information, the photovoltaic array parameters and the inspection time; and by using a heuristic algorithm, based on at least one of the specular reflection area, the photovoltaic array parameters and the expected sampling distance, a determined constraint condition and a cost function, calculating a target shooting point and a target holder attitude of the unmanned aerial vehicle. The accuracy of a photovoltaic inspection result can be improved, and meanwhile, the overall efficiency of photovoltaic inspection is improved.

The application discloses a boundary sample data enhancement method and device for knowledge distillation and a computer storage medium. The method comprises the following steps: before knowledge distillation is performed, the output of a teacher model is used to modify samples in each original data set along the decision boundary of the teacher model step by step, and a plurality of boundary samples suitable for knowledge distillation are expanded. In each iteration, the original sample or each sample modified in the last iteration is used as a basic sample, the approximate tangent plane of the decision boundary near the sample is calculated by using the output of the teacher model, and the sample is modified along multiple directions on the tangent plane; then, the modified sample is modified to be located near the boundary; finally, a plurality of samples farthest from other basic samples are selected as the result of the modification in the round and the basic samples for the next iteration. The application can meet the demand for data enhancement in current image classifier knowledge distillation.

The application discloses a semantic segmentation model training method, a semantic segmentation method, a device and equipment, relates to the technical field of artificial intelligence, and particularly relates to the technical field of deep learning, semantic segmentation, automatic driving and the like. The method comprises the following steps: determining sample point cloud data corresponding to a sample object from original point cloud data based on a preset down-sampling screening mode; the sample object comprises a compulsory object and a to-be-screened object; performing dimension reduction mapping on sample points in the sample point cloud data based on a preset projection rule to obtain sample distance images corresponding to the sample points; and training a semantic segmentation model according to sample label data corresponding to the sample distance images and the sample point cloud data. Through the above technical scheme, the accuracy of semantic segmentation can be improved.

This invention discloses an image feature point description method based on sample mixing and an improved ternary loss function, comprising the following steps: S1, keypoint detection is performed on a given image; S2, image patches of the same size are extracted, centered on each keypoint; S3, the extracted image patches are input into a pre-trained convolutional neural network for image feature description, and the feature descriptor vectors of all keypoints are output. This invention considers the diversity of sample distances and utilizes the classification of sample feature vectors to further increase the accuracy of sample feature description, thereby effectively improving the accuracy of image matching.

A method of coding implemented by a decoding device or encoding device, the method comprising obtaining a value of an angle parameter for a current block; obtaining a value of a width of the current block and a value of a height of the current block; calculating a ratio value between the value of the width and the value of height; obtaining a first value according to the value of angle parameter and the ratio value; calculating a sample distance value for a sample in the current block according to the first value; obtaining a prediction value for the sample in the current block according to the sample distance value.

The SVD algorithm separates a clutter subspace, a target subspace and a noise subspace in radar echoes through singular value decomposition, and the purpose of clutter suppression is achieved. However, due to aliasing of the target and the sea clutter, the clutter suppression effect can be affected by clutter subspace division. In order to solve the problem that the clutter subspace is difficult to determine after the radar echo signal is subjected to SVD decomposition, the invention provides a method for adaptively selecting the clutter subspace by using an improved K-means clustering method. Aiming at the problems that K-means is sensitive to the initial centroid and different characteristic parameters are different in the aspects of statistical stability and anti-noise performance, the initial centroid is fixed, and the sample distance is calculated in a weighted mode. According to the method, after four characteristics of singular value distribution, echo component correlation, Doppler bandwidth and relative Doppler variable coefficient are used as K-means input, clutter subspaces are selected in a self-adaptive mode, and the purpose of clutter suppression is achieved by mapping signals into orthogonal spaces of the clutter subspaces.

The application discloses a kind of X-ray pulsar frequency estimation methods based on swarm optimization, belong to pulsar navigation technical field.The method includes: through waterfall chart folding to observation task, from the consistency of profile signal intensity and segmented phase two aspects to candidate frequency is jointly analyzed;According to the actual sample distribution characteristics reconstruction high-dimensional space sample distance model, nonlinear random neighborhood embedding dimension reduction is carried out to waterfall chart;Three-dimensional image information is converted into two-dimensional plane point group data, and the mean distance origin standard deviation of plane point group is used as the evaluation index of frequency estimation;Combined with the principle of local focusing, the search behavior of scout bees in the bee colony optimization algorithm is improved, and the optimal frequency is efficiently searched;Through simulation comparison with existing methods, the superiority of the method proposed in the application in the performance of pulsar frequency estimation is proved from the aspects of estimation accuracy and operation time.

A method carried out by a computer for generating a 3D model of an object, the method (100) comprising: obtaining a plurality of first 2D images (2A, 2B) which reproduce respective parts of the object (200), obtaining one or more groups of second 2D images (3A, 3B), wherein for each group of second images: the second images of the same group reproduce respective zones of a portion of the object, each second image being at least partially coincident with another second image of the same group, the second images have a respective second ground sample distance which is less than the ground sample distances of the first images, and at least one of the first images and at least one of the second images of the same group reproduce the same element (ei) of the object. For each group of second images, the method comprises: determining a scaling factor between the first images and the second images of the same group on the basis of the element which is reproduced in at least one of the first images and at least one of the second images of this group, resizing the second images of the group of second images by applying the scaling factor and generating a third image (3) which reproduces the portion of the object by merging the second resized images by means of a stitching algorithm which comprises a transformation function which correlates the points of the second resized images with the points of the third image. The method further comprises storing in a data-storage unit (4) the third image of each group of second images in addition to the plurality of first images, and generating a 3D model (5) of the object in a virtual space by means of a photogrammetry technique on the basis of the plurality of first images and the third image of each group of second images.

The invention discloses an extreme weather prediction method and system, and relates to the technical field of artificial intelligence. The method comprises the following steps: collecting multi-modal meteorological observation data; performing space-time alignment and quality correction to obtain an assimilation data set; adjusting a feature extraction weight based on a quality identifier, and projecting the assimilation data set to a low-dimensional hidden space to obtain a hidden space feature vector; based on the sample distance measurement of the hidden space, adopting a meta-learning framework to carry out multi-task prediction on the feature vector of the hidden space; performing multi-target collaborative optimization and antagonism training on the neural network by using the prediction result to obtain a prediction model; and verifying the prediction result by using a digital twin simulation environment to obtain a verification result, and updating model parameters of the prediction model. According to the method, the problem of insufficient prediction precision caused by a complex coupling relationship of meteorological factors and data isomerism is effectively solved, and the accuracy and reliability of extreme weather evolution trend prediction are improved.

A method of coding implemented by a decoding device, comprising obtaining a bitstream; obtaining a value of an indicator for a current block according to the bitstream; obtaining a value of a first parameter for the current block and a value of a second parameter for the current block, according to the value of the indicator and a predefined lookup table; obtaining a value of a sample distance for a sample which is located in the current block, according to the value of the first parameter and the value of the second parameter; obtaining a prediction value for the sample, according to the value of the sample distance for the sample.

The present application relates to a kind of high-throughput testing device and method of electrical properties of component gradient film material, belong to material performance test field, solve the problem that the measurement result is wrong caused by the easy damage of scanning probe to sample.It includes: probe device, located above the sample to be measured, including probe and its fixing device, for detecting the response of sample to external field;Electrical property testing device, for applying external field to sample, collecting the response signal detected by probe, and transmitting the collected data to computer;Displacement device, for loading sample and adjusting the position of sample;Probe-sample distance monitoring device, for real-time monitoring the distance between probe and sample and feeding back to computer;Computer, for controlling displacement device to stop moving at the moment when probe and sample contact, and calculating the electrical properties of sample based on the collected data.The testing device of electrical properties of component gradient film is realized, which can effectively avoid sample damage, fast, efficient and accurate.