68 results about "Rotational invariance" patented technology

The invention provides an unmanned aerial vehicle rapid identification method and system based on a lightweight convolutional neural network. The method comprises the following steps: firstly, acquiring dynamic video stream data and extracting continuous frame images in the dynamic video stream data, then optimizing the continuous frame images to generate optimized continuous frame images, and then inputting the optimized continuous frame images into a lightweight convolutional neural network to generate a feature vector set; performing cross-channel feature fusion on the feature vector set, extracting rotation invariance features and dynamic deformation features, and finally adjusting the parameter response priority of the lightweight convolutional neural network through an adaptive decision module in combination with the real-time transmission rate of the dynamic video stream data to output an identification result. According to the technical scheme provided by the invention, millisecond-level adaptive recognition of the attitude of the unmanned aerial vehicle in a complex environment is realized, and the problems of misjudgment and missing detection of a fixed weight mechanism in a rate fluctuation scene are solved.

The invention belongs to the crossing field of information and ocean science and technology, and discloses a DOA estimation method for a UUV broadside parallel co-prime array. According to the method, firstly, a virtual extension array is generated by utilizing an array auto-covariance and cross-covariance matrix, then, Toeplitz completion is carried out on'holes' in the virtual array through trace norm regularization and nuclear norm regularization constraints, a complete covariance structure is recovered, and finally, an extension matrix is constructed, and rotation invariance of a signal subspace of the extension matrix is utilized, so that the covariance structure of the virtual array is obtained. The pitch angle and the azimuth angle of the target are jointly solved, automatic angle pairing is achieved, and the effectiveness and the reliability of the method are verified through simulation results. According to the method, aiming at the application challenges that the UUV broadside array space is limited and the underwater environment is complex, all virtual array elements of the broadside parallel co-prime array are fully utilized, the array freedom degree and estimation precision are remarkably improved, the target detection capacity of the UUV is effectively enhanced, and the method has high practical engineering application value.

The invention discloses a multi-modal remote sensing image matching method based on modal transformation and comparative learning. The method comprises the following steps: S1, constructing a training data set; s2, constructing a generation path of an image continuous domain; s3, designing a shortest path constraint; s4, defining the total loss of image conversion; s5, extracting rotation invariance feature expression of comparative learning; and S6, establishing a feature matching framework, and outputting a matching result. According to the method, potential sharing features are mined, the generation process of image translation is converted into a step-by-step generation mode based on a network path, the shortest path constraint is realized by constructing an intermediate domain, and the significant nonlinear radiation difference and noise between a reference image and an image to be registered are effectively eliminated; and meanwhile, by utilizing an EfficientNet architecture optimized by an expansion convolution and an attention mechanism, the representation quality of deep features is remarkably improved, and a training system of sample augmentation is designed in combination with twinning and pseudo-twinning networks to obtain better rotation consistency feature expression, so that the matching robustness of a descriptor to a weak texture region is effectively enhanced.

The invention discloses a near-field rotation invariant parameter estimation method based on linear fitting, and the method comprises the steps: calculating the cross-correlation between the receiving data of an array element in a first array and the receiving data of an array element in a second array on a signal source three-dimensional space precise propagation model based on a symmetric cross array, and obtaining a cross-correlation matrix; obtaining a virtual receiving data matrix by vectorizing the cross-correlation matrix; performing eigenvalue decomposition on the covariance matrix of the virtual receiving data matrix to obtain a signal subspace matrix; estimating parameters of a narrowband signal source by using an ESPRIT algorithm, specifically, determining a virtual uniform area array according to a covariance matrix, and partitioning a signal subspace matrix into blocks; the rotation invariance is recovered through the subspace relation of the adjacent block matrixes; obtaining the phase difference between the adjacent sub-arrays through the rotation invariance matrix, building a linear observation model, and obtaining the parameter estimation value of the narrowband signal source; the method has the advantages of low calculation complexity and high universality.

A point cloud denoising method based on a denoising autoencoder. First, the point cloud data is processed, and the point cloud denoising problem is treated as a local problem. The neighborhood of each point is taken and randomly sampled. Secondly, the Transform layer appropriately destroys the input data to create obstacles for subsequent feature extraction. Then, the point cloud is aligned using the rotation matrix calculated by principal component analysis, rotating the point cloud to the same angle. Then, the Encoder layer extracts potential features from the damaged data through a multi-layer perceptron and uses maximum pooling to enhance translation invariance, rotation invariance, and scale invariance. Finally, the Decoder layer of the network decodes the potential features through full convolution and outputs the predicted displacement of the noise point to complete the denoising. The present invention removes noise as efficiently as possible while maintaining the geometric characteristics of the point cloud data.

The invention belongs to the technical field of image processing, and particularly relates to a glass substrate rotating jig alignment method and system based on visual guidance, and the method comprises the steps: constructing a complex gradient vector field according to the gradient of each pixel point in a glass substrate grayscale image, carrying out the four-order phase mapping of the complex gradient vector field, constructing a harmonic transformation field, and carrying out the four-order phase mapping of the harmonic transformation field; calculating a square frame saliency map according to the harmonic transformation field, calculating a circular saliency map based on a complex gradient vector field by using gradient convergence characteristics of solid dots, performing dual fusion on the square frame saliency map and the circular saliency map, constructing a target probability potential energy field, and searching extreme points in the target probability potential energy field to determine true reference point coordinates. And the rotating jig is driven to execute alignment compensation. According to the invention, the interference of metal grid textures, linear scratches and circular stains can be resisted, the rotation invariance is realized, and the high-precision visual alignment is realized.

The present invention discloses a laser SLAM loop detection method based on semantic information, comprising: performing spherical projection on the laser point cloud in the current environment and obtaining the depth information of each point in the two-dimensional image generated after the projection; performing semantic segmentation through a fully convolutional neural network and generating a weight matrix using the obtained semantic labels; aggregating and rotating the weight matrix through a sliding window to generate a global descriptor with rotation invariance; constructing a Kd-Tree to search historical frames and obtain candidate frames based on time thresholds and distance thresholds; obtaining the angular difference between two frames of point clouds using the global descriptor, and performing geometric verification through an ICP algorithm with an initial angle to determine the final loop frame and obtain a closed-loop position. The present invention solves the problem that the existing technology does not fully utilize the semantic information of the environment when performing loop detection using pure laser, resulting in low algorithm efficiency.

This application relates to a fault diagnosis method for abnormal edge trajectory of wind turbine blades. It employs an improved wavelet packet multi-scale analysis of blade operating state images for filtering and noise reduction; it obtains blade trajectory edge features through invariant moment features to achieve translational, scaling, and rotational invariance recognition; and it utilizes a particle swarm optimization algorithm to obtain images with large edge trajectory deviations, calibrating abnormal states for fault diagnosis. This method solves the problems in existing technologies, such as the inability of edge detection operator thresholds to adapt, poor noise resistance, and difficulty in extracting abnormal blade edge trajectory features, leading to ineffective fault diagnosis. It improves fault diagnosis accuracy and, by filtering redundant data through the optimization algorithm, balances the accuracy and efficiency of intelligent recognition, providing reliable technical support for wind turbine blade fault diagnosis.

The invention discloses a geomagnetic gradient tensor depth representation learning method and system, and belongs to the technical field of information processing, and the method comprises the steps: obtaining a geomagnetic three-component data matrix of a magnetic object, and building an original gradient tensor data set; generating a gradient matrix based on the original gradient tensor data set, and performing orthogonal decomposition on the gradient matrix to obtain mutually independent rotation isovariant features and direction sensitive features; inputting the rotation isovariant features and the direction sensitive features into a rotation isovariant convolutional neural network for feature extraction; constructing a geometric constraint loss function including an angle constraint loss function and a direction constraint loss function, and optimizing the training network; the geomagnetic gradient tensor data is subjected to deep representation based on the trained network, the innovative gradient matrix orthogonal decomposition technology solves the problem that rotation invariance and direction sensitivity are difficult to consider at the same time, the feature representation accuracy is improved by about 30-40%, the processing speed is improved by 2-3 times, and the anti-interference capability is remarkably enhanced.

The invention relates to the technical field of artistic pattern recognition, and provides an artistic pattern recognition system based on artificial intelligence, which comprises a data acquisition module, a feature extraction module, a feature selection module, a texture analysis module and a result evaluation module.The artistic pattern recognition system overcomes the defects in the prior art and is reasonable in design and high in practicability. A digital image of the surface of an artwork is obtained through high-resolution digital imaging equipment, denoising, white balance correction and image enhancement processing are carried out to ensure image quality, then texture feature vectors are extracted through a rotation invariant local binary pattern method, texture features can be effectively extracted, rotation invariance is achieved, and classification robustness is improved; the screened features are input into the pre-trained deep learning model, the artistic pattern category probability is output, the deep learning model is used for learning texture features, and the generalization ability and classification performance of the model are improved.

The invention discloses a point cloud semantic segmentation method and system based on rotation invariance and a dynamic weighting mechanism, and the method comprises the steps: obtaining large-scale point cloud data in an automatic driving scene, and obtaining preliminarily purified point cloud data through a density-curvature double-constraint filtering algorithm; acquiring a neighborhood point set of the preliminarily purified point cloud data, splicing and fusing features of the preliminarily purified point cloud data to obtain rotation invariant local features, and forming a coordinate matrix by coordinate information of all neighborhood points in the neighborhood point set; processing the coordinate matrix to obtain structure sensing features; constructing a dynamic graph structure, and processing the structure perception features based on the structure to obtain context enhancement features; statistical information of the rotation invariant local features, the structure perception features and the context enhancement features is processed, and the probability that each point in the preliminarily purified point cloud data belongs to different semantic categories is obtained. According to the method, high-precision segmentation is guaranteed, and meanwhile, the requirement for computing resources can be reduced.

The application discloses a point cloud 3D object detection method not influenced by rotation transformation, comprising the following steps: (1) regarding the network weight of the first layer of the neural network as a vector set distributed in a feature space with the same dimension as the point cloud feature; (2) performing seed point sampling and neighborhood aggregation on the input point cloud data to obtain the local point cloud around each seed point; (3) performing principal component analysis on the network weight and the local point cloud; (4) aligning the weights of the network weight and the local point cloud to obtain a feature with rotation invariance; (5) inputting the local point cloud feature of step (4) into the neural network for feedforward transmission, detecting the prediction of the head output 3D object frame of the network; (6) training the neural network through gradient back propagation; and (7) after the training is completed, performing a 3D object detection task on the point cloud. According to the application, the classification accuracy of the point cloud under arbitrary rotation transformation can be greatly improved, and thus the accuracy of the 3D object detection task can be improved.

The invention discloses a magnetic target attitude measurement method based on gradient tensor. According to the method, gradient tensor measurement information on a measurement surface is utilized, the plane distribution of the equivalent attenuation magnetic moment is obtained through conversion calculation, the local maximum point projection can be regarded as the horizontal position of the target, and the influence of a slowly-varying background magnetic field is eliminated by utilizing the difference characteristic of the gradient tensor; the influence of attitude jitter of the measurement platform is overcome by using the rotation invariance of the equivalent attenuation magnetic moment information, and the projection of the maximum value point of the measurement surface has no systematic deviation; building a quasi-linear equation between the measured value and the target vertical depth and the equivalent magnetic moment by using the equivalent attenuation magnetic moment of the effective measurement point, and further accurately and quickly obtaining the target vertical depth and the equivalent magnetic moment; meanwhile, the possible solution of the equivalent magnetic moment direction of the target is estimated by using the gradient tensor characteristic values and the characteristic vectors of the plurality of effective measurement points, and the real magnetic moment direction of the target is screened out through statistical analysis, so that the target attitude measurement is realized, and the result is steady.

The application provides a two-dimensional DOA estimation method for arbitrary array single-base MIMO radar based on data rearrangement, and relates to the technical field of array sensor direction finding. The rank structure of a coherent signal covariance matrix is recovered by rearranging a received data matrix through arbitrary array element positions of a transmitting array and a receiving array; S2: eigenvalue decomposition is performed on the signal covariance matrix to obtain a signal subspace; S3: reliable rough estimation is obtained through rotation invariance; S4: an integer ambiguity vector corresponding to spatial phase ambiguity caused by array element spacing greater than half a wavelength is solved; S5: the phase of an array flow pattern with phase ambiguity is compensated, and an accurate estimation value of the 2D-DOA of the target is calculated; the high-precision 2D-DOA estimation problem in a complex scene of a coherent signal source, arbitrary array geometry and phase ambiguity caused by part of array element spacing greater than half a wavelength is solved, unambiguous and high-precision angle parameter estimation is realized, and the estimation accuracy and algorithm reliability are improved.

The application discloses a geomagnetic gradient tensor deep characterization learning method and system, and belongs to the technical field of information processing. The method comprises the following steps: acquiring a geomagnetic three-component data matrix of a magnetic object and establishing an original gradient tensor data set; generating a gradient matrix based on the original gradient tensor data set, orthogonally decomposing the gradient matrix, and obtaining mutually independent rotation invariance features and direction sensitive features; inputting the rotation invariance features and the direction sensitive features into a rotation invariance convolutional neural network for feature extraction; constructing a geometric constraint loss function comprising an angle constraint loss function and a direction constraint loss function, and optimizing a training network; and performing deep characterization on geomagnetic gradient tensor data based on the trained network. The innovative gradient matrix orthogonal decomposition technology solves the problem that rotation invariance and direction sensitivity cannot be considered simultaneously, the feature characterization accuracy is improved by about 30-40%, the processing speed is improved by 2-3 times, and the anti-interference capability is significantly enhanced.

The application discloses a kind of fast and multi-modal image matching method with scale and rotation invariance, comprising the following steps: step 1, local intensity binary conversion, step 2, FAST feature detection and projection-based scale space construction, step 3, rotate local image block and use SIFT-like descriptor to describe feature points, step 4, with Euclidean distance as the measure, use the nearest neighbor distance ratio-free brute force matching algorithm to match.The application first proposes a kind of local intensity binary conversion model, the similarity between multi-modal images after the conversion is significantly improved, can effectively reduce the nonlinear radiation difference between multi-modal images.Meanwhile the application also proposes a new scale space construction method, i.e.projection-based scale space, not only realizes scale and rotation invariance, but also can realize fast high-precision matching on multi-type large-size multi-modal images, has higher application value and wide application prospect.

The invention discloses a deep watermarking technology robust to any rotation angle, and aims to solve the problem that the robustness of digital watermarking is reduced due to rotation attack in an image transmission process. The method comprises an encoder module which is used for embedding secret information into a host image to generate a watermark-containing image; the double-noise-layer strategy comprises a rotation noise layer and a conventional noise layer, and the robustness to rotation attack and conventional attack is synchronously improved through adversarial training; the decoder module is used for integrating a rotation feature normalization device and a deformable convolutional network and enhancing rotation invariance through multi-angle feature aggregation and adaptive spatial sampling; the loss function is used for optimizing watermark invisibility and extraction accuracy in combination with embedding loss, decoding loss and adversarial loss. Compared with the existing deep watermarking technology, the method provided by the invention has the advantages that the characteristics of the rotation feature normalization device and the deformable convolutional network are fully utilized, the high watermarking extraction accuracy at any rotation angle is realized, and meanwhile, the high visual fidelity is kept.

The invention relates to a wind turbine generator blade edge trajectory abnormal fault diagnosis method. The method comprises the following steps: performing multi-scale analysis on a blade operation state image by adopting an improved wavelet packet to perform filtering and noise reduction; blade track edge features are obtained through invariant moment features, and translation, stretching and rotation invariance recognition is achieved; an image with large edge trajectory deviation is obtained by using a particle swarm optimization algorithm, and an abnormal state is calibrated for fault diagnosis; the method solves the problems that in the prior art, an edge detection operator threshold cannot be self-adaptive, noise resistance is poor, and faults cannot be effectively diagnosed due to the fact that blade edge track abnormal features are difficult to extract. And redundant data are screened out through an optimization algorithm, the accuracy and efficiency of intelligent identification are considered, and reliable technical support is provided for fault diagnosis of the wind turbine generator blades.

The invention relates to the technical field of unmanned aerial vehicle detection, in particular to a low-altitude unmanned aerial vehicle target detection method based on spectrum anomaly recognition, which comprises the following steps: S1, collecting radio frequency signals of a monitoring area, and generating a plurality of time domain sampling sequences; s2, performing complex number self-correlation operation on the time domain sampling sequence, and extracting dynamic characteristics with rotation invariance; and S3, determining the target existence of the unmanned aerial vehicle according to the trajectory closing index of the dynamic characteristics in the phase space. According to the method, passive, non-contact and high-precision detection of a low-altitude unmanned aerial vehicle target is achieved, compared with a traditional energy detection or frequency spectrum scanning method, rotor signals can still be stably extracted in a non-line-of-sight, multi-path reflection or lightning noise environment, the closed structure of a rotor track is effectively recognized, and the detection accuracy is improved. And the problem that the traditional'head and tail point distance 'is easy to misjudge is avoided.

The application relates to the technical field of image recognition, in particular to a target recognition method and device, electronic equipment and a storage medium. The application obtains a plurality of transformed feature maps by performing a plurality of affine transformations on a first feature map of a first frame of image, determines a target rotation angle through rotation invariance scores between each transformed feature map and a second feature map, and then inputs the transformed feature map corresponding to the target rotation angle into a trained target rotation equivalence encoder to obtain a template feature vector. Further, based on the rotation invariance score between a query feature vector of a second frame of image and the template feature vector, it is determined whether a candidate object in the second frame of image is a target object. In this way, continuous recognition of a target object performing three-dimensional rotation in a target video can be realized, and the accuracy of target object recognition can be improved.

The invention provides a multi-modal image matching method for scene matching navigation, which relates to the technical field of image matching, and comprises the following steps of: filtering a first modal image and a second modal image to obtain even symmetry and odd symmetry of each pixel point, further obtaining the maximum moment of each pixel point, and further obtaining boundary information; and further obtaining feature points, obtaining feature descriptors of the feature points, and performing feature matching to obtain a final matching result of the first modal image and the second modal image. The method is insensitive to radiation distortion, has rotation invariance, and solves the matching problem caused by non-linear radiation distortion of different-source images.

The application discloses a point cloud scene semantic instance joint segmentation method based on adaptive feature fusion, and specifically comprises the following steps: constructing an LPR-AFFN network; constructing a local polar representation module LPR, and performing LPR operation on the original point cloud; performing coordinate conversion according to Z-axis rotation invariance, calculating a neighborhood centroid point matrix, updating a polar angle, and combining the polar coordinates, geometric distances and original point cloud information to obtain a feature matrix; constructing a feature representation module, inputting the feature matrix into the feature representation module, defining a local centroid point, finding the neighborhood points of the centroid point to construct a local group; preliminarily performing feature extraction on the input point cloud; constructing a feature extraction module, obtaining new discriminative features after the extracted features pass through the feature extraction module; and constructing an adaptive feature fusion module, and using the adaptive feature fusion module to integrate the new discriminative feature information to construct the global perception ability of the point cloud.

The present invention discloses a multi-array non-circular source rapid positioning method based on ESPRIT and weighted dimensionality reduction search, comprising the following steps: first, using the elliptical covariance information of the target signal to expand the spatial information to obtain an increased virtual array aperture; second, obtaining the azimuth information of each observation station through the rotational invariance technique (ESPRIT); then, using the non-circular phase to associate the azimuth information of each observation station with the signal source; and then, combining the information of all base stations and using the least squares method to directly solve the target position as an initial estimate. Finally, a weighted dimensionality reduction search is performed within a small range near the initial estimate to improve the estimation accuracy. Compared with traditional two-step positioning algorithms, subspace data fusion algorithms (SDF), and Capon direct positioning algorithms, the present invention has higher spatial degrees of freedom and positioning accuracy, and can estimate more targets. In addition, the method significantly reduces the computational complexity while ensuring estimation performance.

The invention discloses a target object track movement mode recognition method, device and equipment and a medium, and relates to the technical field of data analysis, and the method comprises the steps: obtaining the track data of a target object, carrying out the linear interpolation and low-pass filtering smoothing processing of the track data, and obtaining a track sequence; calculating an instantaneous displacement direction and a discrete curvature based on the trajectory sequence, and determining direction quadrant information and trajectory curvature information of the trajectory sequence by using the instantaneous displacement direction and the discrete curvature; mapping and compressing the trajectory sequence direction quadrant information and the trajectory curvature information to generate a character sequence; constructing sequence templates of different track types, and comparing the sequence templates of different track types with the character sequence; the sequence template is a template representing a track moving mode of the target object; and the track movement mode of the target object is identified based on the comparison result, multiple target object track movement modes with movement and rotation invariance can be identified, and the efficiency and universality of track movement mode identification are improved.

The application discloses a kind of generation methods of rotation invariant multi-scale annular feature descriptor, to overcome the challenge that feature recognition technology is difficult to accurately describe key features due to image rotation problem in image recognition field.First, multi-scale annular domain is constructed, for the segmentation local feature around feature point, to obtain multi-scale annular feature;Then the eigenvalue of annular feature is calculated with annular feature as basic unit;Finally, the eigenvalue obtained from multiple scales is sorted to form a feature vector with rotation invariance.The feature description scheme provided by the application avoids obtaining a feature vector with rotation invariance by estimating the dominant direction of the feature point.Compared with the scheme based on the dominant direction, the feature vector provided by the application will not be disturbed by the error from the dominant direction, avoids the operation of rotating all local features to the same direction according to the dominant direction, simplifies the calculation steps of describing local features, and reduces the algorithm complexity.

The application discloses a two-stage heterogeneous remote sensing image registration method with radiation and rotation invariance, comprising the following steps: performing WLD filtering on a group of images to be registered to obtain a WLD structure saliency map and a WLD direction map sequence of the group of images to be registered; constructing a feature description vector of the group of images to be registered based on the WLD structure saliency map and the WLD direction map sequence; matching feature points between the group of images to be registered according to the feature description vector to obtain a rotation reference point pair; extracting a WLD structure feature description vector of the group of images to be registered according to the rotation reference point pair, and matching feature points between the group of images to be registered again according to the WLD structure feature description vector to obtain a plurality of pairs of final matching point pairs; and registering the group of images to be registered according to the final matching point pairs. The application can enhance registration accuracy and effect.

The application discloses a kind of based on graph semantic redirection module's remote sensing image ship target detection method and system, comprising: input remote sensing image to target detection model, carry out feature extraction by stacked convolution block, obtain shallow, middle and deep feature map and utilize feature pyramid network to carry out feature fusion;The improved detection head of feature map after fusion is introduced into graph semantic redirection module, graph semantic inference is carried out by introducing multilayer perceptron, pixel-level semantic similarity is used to construct graph structure to extract key features with rotation invariance, global semantic consistency integration is carried out using the feature integration mechanism based on exponential moving average, decoupling feature map obtains target class and position information;Target detection model is trained and tested for target detection task, and ship target detection is carried out on the remote sensing image to be measured;The application realizes task feature decoupling in detection head stage and enhances the extraction of rotation invariance semantic features, so as to realize high-precision remote sensing ship detection.

The present application discloses a data processing method, device, apparatus, and computer storage medium. The data processing method disclosed in the present application includes: processing an image to be processed to obtain first data; wherein the first data represents pulse data corresponding to the image to be processed; performing feature extraction and pooling operations on the first data through multiple pulse neurons carrying synaptic weight information in at least one feature extraction layer to obtain first feature data; processing the first feature data to obtain second data; wherein the second data represents the type of object in the image to be processed. The data processing method provided in the present application can improve the accuracy of object recognition in the image to be processed, and can also overcome the disadvantage of convolutional neural networks lacking rotation invariance.

The present invention relates to the technical field of target recognition, and particularly provides an aircraft target recognition method based on circular filtering and convolutional neural network. By using the shape and structure characteristics of the aircraft itself, circular filtering features are extracted for feature enhancement, combined with the features of the deep neural network, and finally, aircraft target recognition based on circular filtering and convolutional neural network is learned. The present invention has rotational invariance and scale invariance, and has more accurate positioning compared with the recognition using only convolutional neural network, can accurately recognize aircraft targets, and has a high recognition accuracy and recall rate.

Point cloud identification device (300), comprising: a point cloud acquisition unit (320) to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit (310) to acquire a model that has a learning parameter; a rotation invariance conversion unit (330) to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotation invariance feature using data after orthogonalization; an inference unit (340) to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; and a result output unit (350) to output a classification result by identification by the inference unit (340).