38 results about "Translation invariance" patented technology

The invention belongs to the technical field of signal processing, and particularly relates to an under-snapshot 2D-MUSIC direction finding method based on spatial translation invariance. According to the method, the space translation invariant characteristic of a uniform linear array or a uniform planar array is utilized, a calculation method of a forward and backward space-time cross-correlation function is provided, a pseudo covariance matrix is constructed by using the space-time cross-correlation function based on the Hermitian-block Toeplitz characteristic of a signal ideal covariance matrix, then a MUSIC spatial spectrum is constructed, and the wave direction is extracted. Compared with a traditional 2D-MUSIC algorithm, the method still can maintain the high-precision angle estimation capability under the condition that the number of snapshots is limited, and does not greatly improve the calculation amount. Compared with a spatial smoothing MUSIC algorithm, the method does not need to divide sub-arrays, fully utilizes the full array aperture, and avoids the loss of resolution.

The invention specifically relates to a method and a system for detecting a high-resistance grounding fault of a power distribution network, and belongs to the technical field of power distribution network fault detection. According to the method, the amplitude difference of the zero-sequence current is converted into the angle difference through the Grubby angle field coding, so that weak current distortion is amplified into a remarkable spatial stripe in the Grubby angle field image, and the abrupt change characteristic in the fault transient process can be effectively represented. Meanwhile, due to the symmetry and translation invariance of the Grubrum angle field matrix, the time domain dependency relationship of the zero-sequence current waveform of the high-resistance grounding fault is completely reserved, and the fault feature extraction capability can be effectively improved. According to the method, a traditional standard convolution kernel is improved according to polar coordinate characteristics of the Grubrum angle field image, time evolution characteristics are extracted by using radial branches of the multi-axis variable neural network, amplitude distortion characteristics are extracted by using angular branches of the multi-axis variable neural network, diagonal characteristics are extracted by using deformable convolution in a self-adaptive manner, and the expression ability of the extracted characteristics is effectively improved.

The invention discloses an aircraft high lift device two-dimensional to three-dimensional optimization method based on deep reinforcement learning, and belongs to the technical field of aircrafts. According to the method, in an environment based on two-dimensional computational fluid mechanics, an efficient and universal aircraft high lift device optimization strategy suitable for a target three-dimensional optimization problem is trained, the training cost of the three-dimensional optimization problem is remarkably reduced, and high calculation overhead of traditional three-dimensional flow field simulation is avoided; the flow field locality and strategy translation invariance are utilized to decompose a three-dimensional problem into a plurality of two-dimensional profiles for independent optimization, so that an intelligent agent only needs to process a low-dimensional action space, and the problem of curse of dimensionality caused by a high-dimensional action space is effectively solved; when the trained intelligent agent is used for solving the similar optimization problem, the optimization efficiency is high, the convergence speed is high, and the method has high universality for variable design working conditions and the appearance of the high lift device by means of the previous experience for solving the similar problem.

The invention discloses a visual Transform-based feature focusing image semantic segmentation method, which adopts a feature pyramid structure to generate four feature maps with different sizes, and uses an overlapping block embedding module in a feature map embedding process to ensure that local information is continuous to a certain extent while the feature maps are partitioned. A position code generation module is used for dynamically generating corresponding position codes according to local neighborhood information, translation invariance needed in computer vision is kept, and then a fusion module combining a focusing attention mechanism and convolution operation is adopted in the Transform coding process. A self-attention mechanism is utilized to grasp a dependency relationship between an overall structure and a long distance, and convolution attention is introduced to enable the model to maintain sensitivity to local information, so that the model can integrate global and local detail information at the same time. And finally, inputting the feature maps of different sizes generated in each stage of the model into a decoder for segmentation. According to the method, the segmentation performance of a target boundary and a complex scene is improved.

A method for place re-recognition of a mobile robot based on a lidar estimable pose, said method comprising: using radon transform to convert rotation and translation changes into translation changes on two axes of a sinusoidal graph, and, on the basis of an amplitude spectrum of a spectrum, performing spectrum cross-correlation calculation on translation invariance and two images, so as to solve the translation property of the images. Translation invariance is used to generate a position descriptor and thereby perform candidate matching for place re-recognition; and cross-correlation calculation may be performed together with radon transform to solve relative rotation and translation. In the described method, a time-varying environment is considered, and by using a multi-channel feature BEV for representation, the capability of representing a local feature in a laser point cloud can be improved.

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 present invention discloses an intelligent mineral sorting method based on convolutional neural networks and self-attention mechanisms, which realizes intelligent mineral sorting with high precision, high efficiency, low volume and low computational complexity. This architecture combines the inductive bias of locality and translation invariance in convolution operations with the globality and long-distance dependence of the self-attention mechanism to establish a mineral image classification model with stronger recognition ability and better feature capture ability. Compared with a single network based on convolutional neural networks or self-attention mechanisms, the architecture proposed by the present invention has higher classification accuracy and lower computational complexity. Secondly, the architecture provided by the present invention is suitable for fine-grained classification tasks of multiple types and categories of mineral images, and has excellent discrimination ability for mineral images with small differences in apparent features.

The invention discloses a DNAPLs inversion method based on Swin-Transform fusion multi-source data, and the method comprises the steps: generating a saturation field number and a permeability coefficient field array of DNAPLs, and forming an input tensor; preprocessing the input tensor to obtain a preprocessed tensor; the preprocessed tensor is input into a Swin Transform backbone network for feature extraction, and a feature tensor is obtained; re-parameterization is carried out to obtain assimilation hidden vectors; obtaining a first assimilation reconstruction tensor; judging the number of iterations to obtain a multi-source simulation observation vector of the first assimilation; and marking each channel of the finally obtained assimilation reconstruction tensor as a saturation field array and a permeability coefficient field array. According to the method, spatial position codes consistent with channel dimensions are introduced, the expression ability of a non-stationary space structure is improved, spatial position information is explicitly injected, responses of different spatial positions can be distinguished in the subsequent Swinin-Transform feature extraction process, and the limitation that only convolution translation invariance is relied on is avoided.

The application discloses a SAR polar coordinate imaging method and device based on flat flight equivalence, and the method comprises the following steps: constructing a flat flight equivalence slant range model of a large oblique SAR to obtain a baseband echo signal formed by target area scattering; performing distance direction matched filter processing on the baseband echo signal to obtain a signal after range pulse compression; sequentially performing Dechirp processing, two-dimensional resampling processing and IFFT transformation on the signal after range pulse compression to obtain a SAR polar coordinate primary imaging result; and performing geometric correction on the primary imaging result based on reverse projection to obtain an imaging result without geometric deformation and with good focusing. The SAR polar coordinate imaging method based on flat flight equivalence provided by the embodiment avoids the problem of azimuth translation invariance caused by diving through slant conversion, solves the problem that the traditional PFA is not applicable to diving SAR, and successfully extends the PFA imaging algorithm without deformation to diving SAR by correcting the geometric deformation problem caused by diving.

The invention provides an urban rail vehicle-mounted autonomous obstacle detection method and system based on millimeter wave radar map convolution, and belongs to the technical field of rail traffic obstacle detection. Millimeter wave radar is initialized, and collected radar frames are continuously input into a queue; reading a plurality of frames of radar point cloud data from the queue, and converting the processed point cloud data into a graph structure; and processing the graph structure by using the trained graph convolutional neural network, finally outputting an obstacle detection and classification result, and performing early warning on train driving. According to the invention, high-dimensional embedding helps to realize information recombination and high-dimensional aggregation of channel features, and rotation and translation invariance ensures the accuracy and robustness of obstacle detection; the jump connection structure effectively solves the problem of excessive smoothness of the graph neural network; through a unique network structure and feature selection and division, the feature relationship between different points is concerned, global information is acquired in combination with an attention mechanism, and more efficient and accurate rail transit obstacle classification and detection are realized.

The invention discloses a frequency spectrum recognition training method for translation invariant adversarial attacks, which belongs to the field of image training, and comprises the following steps: constructing a translation invariant attack target function, and generating adversarial samples for a group of pictures subjected to translation transformation; the translation invariant attack method can effectively reduce the degree of dependence of the adversarial sample on the adopted white box model, thereby improving the migration attack success rate of the black box defense model. In order to more efficiently optimize the translation invariant attack target function, under the assumption that the convolutional neural network has translation invariance, an attacker can calculate the gradient of a model loss function for a non-translated picture, and then the gradient and a preset kernel matrix are subjected to convolution to approximate the gradient of the translation invariant attack target function. The translation invariant attack method can be combined with all gradient-based attack methods, and the experimental result verifies the effectiveness of the proposed method.

The application discloses a kind of mobile robot high-efficiency robust global positioning method based on laser radar.The application uses Radon transformation to convert rotation and translation change to the translation change of two axes of sinogram, uses translation isovariant feature extraction network to carry out feature extraction, guarantee rotation and translation isovariance, using Fourier transform to obtain the amplitude spectrum of frequency spectrum to realize translation invariance, by cross-correlation operation realizes rotation translation invariance similarity calculation, while supervising network extracts feature suitable for place identification task, to improve the representation ability of rotation translation invariant feature in laser point cloud.In addition, the application uses cross-correlation operation to estimate relative rotation and translation to provide good initial value for point cloud registration algorithm, and further solve accurate 6 degrees of freedom relative pose.

The application discloses a kind of image block division pre-processing methods of visual Transformer, the method is for the locality of visual Transformer, translation invariance lacking to image, PVT image classification algorithm is improved, on the basis thereof, image block has been treated with multi-scale feature fusion, multi-scale feature fusion image block embedding layer (MSFF_PE) is proposed, through the lightweight multi-scale feature fusion module designed, the multi-scale feature information in image block is enriched, and the feature representation capability of image block is strengthened.For the modeling of lacking local information in the structure of encoder, a multi-scale local enhancement feedforward neural network layer (MSLE_FF) is proposed, the one-dimensional sequence output by the attention mechanism is restored to a two-dimensional image for local information interaction, and the weighted information of local features is obtained by introducing distance importance. Finally, the effectiveness of the method of the application is verified through experiments.

The invention provides a multi-sound-source detection and positioning method and system of a convolutional recurrent neural network. Amplitude and phase features of multi-channel audio are extracted, an embedded layer is used for generating a position code for each frame of audio, the features and the position codes are jointly input into a neural network, and the convolutional neural network is used in a network structure to learn distinguishing of sound source categories and sound source positioning according to features between channels. Features are extracted from different scales by using multiple groups of void spatial pyramid pooling among convolutional layers, translation invariance of a convolutional neural network is kept by using a spatial converter network, context information is learned by using a gating loop unit based on an attention mechanism, and the sound source position of the current frame is predicted in an auxiliary manner according to the information of the previous frame; global features are obtained through global average pooling processing, the global features and the position codes are combined and input into a parallel full connection layer, and a sound event prediction result and an arrival direction prediction result are output. According to the invention, respective positioning of simultaneous sounding of a plurality of sound sources is realized, and robustness is provided for reverberation and low signal-to-noise ratio environments.

A kind of off-grid photovoltaic array fault diagnosis method based on ConvTran, comprising the following steps: constructing Transformer model;Convolution module is used on the Transformer architecture, before input embedding vector is input into Transformer module, position embedding vector generated by tAPE is added to input embedding vector, after obtaining the final output of Transformer module, global average pooling and full connection layer are applied to process, obtain more translation invariance model, finally, the result of classification prediction is obtained by applying Softmax function;Collect the multivariate time series data of four channels of photovoltaic array, input ConvTran network model, and carry out fault diagnosis to photovoltaic array.The method is more suitable for the fault type classification of multivariate time series, and shows more excellent applicability and accuracy in off-grid photovoltaic array fault diagnosis.

The invention belongs to the technical field of signal processing, and particularly relates to an under-snapshot MUSIC direction finding method based on spatial translation invariance. According to the method, a space translation invariant characteristic of a uniform linear array is utilized, a calculation method of a space-time cross-correlation function is provided, a pseudo-covariance matrix is constructed by using the space-time cross-correlation function based on a Hermitian-Toeplitz characteristic of a signal ideal covariance matrix, a MUSIC spatial spectrum is constructed, and a wave direction is extracted. Compared with a traditional MUSIC algorithm, the method still can maintain the high-precision angle estimation capability under the condition that the number of snapshots is limited, and does not greatly improve the calculation amount. Compared with a spatial smoothing MUSIC algorithm, the method does not need to divide sub-arrays, fully utilizes the full array aperture, and avoids the loss of resolution.

The present application provides a seismic phase classification method, device, electronic device and medium. First, a two-dimensional scattering transform is used to extract multi-scale feature information with local deformation stability and translation invariance from two-dimensional seismic data. That is, when the seismic signal is affected by underground fault activity and produces local micro-deformations and time shifts, the extracted multi-scale features will not change significantly, which is conducive to the accurate classification of seismic data. Then, a densely connected convolutional network is constructed based on DenseBlock, and convolutional neural networks of different depths are used to further process and fuse feature information of different scales, that is, multi-resolution processing, and finally realize seismic phase classification of seismic data. Since DenseBlock adopts a fully interconnected structure, feature reuse can be achieved, so that feature information can be better retained during the layer-by-layer processing, further improving the accuracy of seismic phase classification results.

This invention discloses a channel estimation method for a flexible antenna array-assisted ISAC system. It constructs a multidimensional tensor channel model that integrates antenna rotation parameters. The received signals are then stacked into a tensor form with a Vandermonde structure. A two-stage parameter estimation framework is employed: the first stage utilizes the translation invariance of the tensor in the time delay dimension to achieve initial estimation of time delay and Doppler shift using the ESPRIT algorithm; the second stage performs tensor normalized multilinear decomposition within a Bayesian probabilistic framework, introduces hierarchical sparse priors to adaptively determine the model rank, and performs joint fine estimation of multidimensional channel parameters such as angle of arrival, departure angle, time delay, Doppler shift, and reflection coefficient. Simulation results show that this invention has higher estimation accuracy and stability under complex channel and low signal-to-noise ratio conditions, and is suitable for broadband large-scale ISAC systems assisted by flexible antenna arrays.

The application provides a kind of city rail vehicle autonomous obstacle detection method and system based on millimeter wave radar graph convolution, belongs to the technical field of rail transit obstacle detection, millimeter wave radar initialization, constantly input collected radar frame to queue;Read several frames of radar point cloud data from the queue, convert the processed point cloud data into a graph structure;The trained graph convolutional neural network is used to process the graph structure, and finally outputs the obstacle detection and classification result, and makes a warning for train travel.The high-dimensional embedding helps to realize information reorganization and high-dimensional aggregation of channel characteristics, and the rotation and translation invariance ensures the accuracy and robustness of obstacle detection;The jump connection structure effectively solves the over-smoothing problem of graph neural network;Through unique network structure and feature selection division, the feature relationship between different points is focused, and global information is obtained by combining attention mechanism, to realize more efficient and accurate detection of rail transit obstacle classification and detection.

The invention discloses an automatic adjusting method and system for an insulin pump, and relates to the technical field of automatic infusion control, and the method comprises the steps: collecting a signal output by a sensor in a motion monitoring unit, and preprocessing the signal to obtain a preprocessed signal frame; performing wavelet scattering transform processing on the preprocessed signal frame to generate a multi-order scattering coefficient set; calculating statistical characteristic parameters and constructing a multi-dimensional scattering characteristic vector; inputting the multi-dimensional scattering feature vector into a lightweight time sequence convolution class network, and outputting an event classification result; performing adaptive adjustment control on the infusion state of the insulin pump based on the event classification result; through translation invariance and multi-scale decomposition capability of wavelet scattering transformation and in combination with self-adaptive learning capability of a lightweight time sequence convolution network, high-precision real-time distinguishing of real mechanical shock and electromagnetic coupling interference is realized on an embedded platform, the false alarm rate in an electromagnetic environment is remarkably reduced, and the reliability of the system is improved. And infusion continuity and time sequence integrity are guaranteed.

The present application belongs to the technical field of signal processing, and particularly relates to a kind of under fast 2D-MUSIC direction finding method based on spatial translation invariance.The method of the present application utilizes the spatial translation invariance characteristics of uniform linear array or uniform planar array, proposes the calculation method of forward and backward space-time cross-correlation function, and then based on the Hermitian-block Toeplitz characteristics of signal ideal covariance matrix, constructs pseudo covariance matrix using space-time cross-correlation function, and further constructs MUSIC spatial spectrum to extract the direction of arrival.Compared with the traditional 2D-MUSIC algorithm, the present application can still maintain high-precision angle estimation ability under the condition of limited fast shots, and does not significantly increase the computational load;Compared with the spatial smoothing MUSIC algorithm, the present application does not need to divide subarray, fully utilizes the full array aperture, and avoids the loss of resolution.

The application discloses a point cloud registration method and system based on a rotation equivariant network, and the method comprises the following steps: obtaining point clouds to be registered; processing each point cloud to be registered by using a first network based on point rotation equivariance and translation invariance, obtaining a reference direction and a key point score of each point cloud; extracting key points according to the key point score, and extracting a rotation invariant and SO(2) equivariant feature for each key point according to the reference direction and a second network based on patches; based on a preset matching strategy, generating point correspondences with a high inlier rate according to the features; and solving a relative pose by using a corresponding grouping algorithm according to the point correspondences, thereby completing point cloud registration. The embodiment of the application constructs an efficient and universal feature learning framework, can generate point correspondences with a high inlier rate by using rotation invariant and SO(2) equivariant features, and can be widely applied to the technical field of computers.

Provided is a fault signal locating and identifying method of industrial equipment based on a microphone array. The method includes the steps of: acquiring sound signals and dividing the acquired signals into a training set, a verifying set and a test set; performing feature extraction on the sound signals in the training set, and extracting a phase spectrogram and an amplitude spectrogram of a spectrogram; sending an output of a feature extraction module, as an input, to a CNN, and in each layer of the CNN, learning a translation invariance in the spectrogram by using a 2D CNN; in between the layers of the CNN, normalizing the output by using a batch normalization, and reducing a dimension by using a maximum pooling layer along a frequency axis; sending an output from the layers of the CNN to layers of RNN; using a linear activation function; and inputting an output of a full connection layer to two parallel full connection layer branches for fault identification and fault location, respectively.

The invention provides a regional independence target attack confrontation sample generation method, device and equipment, and relates to the technical field of deep learning security. The method comprises the following steps: acquiring an input original image; a first mask and a second mask are randomly and dynamically generated based on the original image, the second mask is a complementary area of the first mask, and the second mask and the first mask are not overlapped in space; taking the first mask and the second mask as two independent disturbance areas during disturbance optimization, and respectively applying the two independent disturbance areas to the original image to generate two independent sub-adversarial samples; inputting the two sub-adversarial samples into a convolutional neural network to carry out gradient calculation and optimization, and carrying out gradient updating in combination with translation invariance and input transformation diversity so as to update a disturbance region; and adding the optimized disturbance region to the original image to generate a new adversarial sample, wherein the new adversarial sample is used for misleading the deep learning model to output a specified error category. According to the method, the mobility and the stability of the target attack are remarkably improved, and particularly, the performance is excellent in a black box scene.

The application discloses a pathological section tumor stroma ratio quantification method, and steps are as follows: (1) a weakly supervised neural network training process based on an image block: firstly, block a full section image scanned by a digital scanner, then assign a unique label, i.e., tumor or non-tumor, to each image block, and finally, train a neural network added with a two-dimensional random inactivation layer using cross-entropy loss and translation invariance loss; (2) a tumor stroma ratio quantification algorithm based on a neural network: generate a class activation mapping graph using a previously trained model to identify a section, calculate a stroma region using a morphological algorithm, and further calculate a tumor stroma ratio. The application is suitable for making an accurate quantitative judgment on the tumor state of a patient in a clinic, and compared with existing methods, the application has the characteristics of high repeatability, low execution cost and high calculation accuracy, can help a pathologist to quantize a prognosis factor difficult to manually calculate, and significantly reduces the workload of the pathologist.

The invention discloses an aero-engine sensor multi-fault self-diagnosis method based on a wavelet scattering network, and aims to solve the problem of insufficient diagnosis accuracy under the condition that multiple faults coexist and computing resources are limited. According to the method, firstly, sensor signals are collected, and a training data set covering normality, abrupt change, drift, bias and periodic disturbance is constructed; then, deep time-frequency features with translation invariance are extracted by using a wavelet scattering network, and signal energy distribution is effectively captured; dimensionality reduction is carried out on the high-dimensional features through principal component analysis, and redundant information is removed; and finally, a support vector machine multi-classification model is constructed based on the dimension reduction features, and rapid identification of various faults is realized. The method has high precision and high real-time performance under the condition of finite computing power, and the fault diagnosis capability of an aero-engine sensor system is remarkably improved.

The invention provides an air combat target maneuvering multi-modal trajectory prediction method and system based on a space-time joint attention mechanism. The method comprises the following steps: modeling an air combat scene; modeling translation invariance and rotation invariance; carrying out stability modeling; modeling space-time correlation; performing multi-mode decoding; and verifying feasibility. According to the method, synchronous capture of time mutation and spatial game is realized through a double-branch attention mechanism; multi-modal prediction is used to generate a plurality of possible trajectories for a target in a heterogeneous complex high-dynamic environment, and uncertainty and variability in trajectory prediction are considered; an aerodynamic penalty term is introduced into a loss function, it is ensured that the trajectory conforms to the warplane body limit, physical distortion is eliminated, and the accuracy and reliability of trajectory prediction are further improved.

A method for place re-recognition of a mobile robot based on a lidar estimable pose, said method comprising: using radon transform to convert rotation and translation changes into translation changes on two axes of a sinusoidal graph, and, on the basis of an amplitude spectrum of a spectrum, performing spectrum cross-correlation calculation on translation invariance and two images, so as to solve the translation property of the images. Translation invariance is used to generate a position descriptor and thereby perform candidate matching for place re-recognition; and cross-correlation calculation may be performed together with radon transform to solve relative rotation and translation. In the described method, a time-varying environment is considered, and by using a multi-channel feature BEV for representation, the capability of representing a local feature in a laser point cloud can be improved.

The invention discloses a weighted ternary feature loopback detection method for laser SLAM (Simultaneous Localization and Mapping), and relates to the technical field of positioning and navigation. The invention provides a global descriptor of weighted ternary features, and the global descriptor has rotation and translation invariance; firstly, plane points and edge points which are easy to distinguish are screened out through geometric feature constraints; then, a local self-adaptive coordinate system is constructed on the basis of the feature points and the main direction of the feature points and used for extracting stable ternary features with physical significance, and parameterized expression of the point pair relation is achieved; and finally, fusing normal difference and curvature difference to construct a weighting mechanism, reinforcing geometric description of a local field, introducing density weight to obtain frame-level descriptors with global expression capability, and using Euclidean distance between the descriptors as an inter-frame similarity measurement standard to judge whether loopback exists or not. An experiment result on a public data set KITTI shows that compared with LeGO-LOAM, the root mean square error is reduced by 32% at most.

The present invention relates to a matching method, device, and storage medium for significantly rotated images. The method comprises the following steps: inputting a reference image and a target image and performing preprocessing; based on the inherent translation invariance of a convolutional neural network, simplifying the rotated image matching problem to a corresponding point matching problem using a group equivariant neural network modeled based on the plane rotation group SO2; constructing and training a rotation equivariant image feature extractor; establishing two matching routes, and applying a nearest neighbor matching algorithm to the image features of the images in each matching route; and performing a primary filtering, backprojection, merging, and secondary filtering on the matching results of the two matching routes using a dual rotation equivariant excitation mixer, to output a matching relationship of dense corresponding points in the reference and target images. Compared with the prior art, the present invention can overcome the effects of significant illumination changes and achieve image matching under all rotation angles.