17results about How to "Implement feature extraction" patented technology

The invention discloses a three-dimensional heterogeneous integrated optical computing chip, and belongs to the field of optical computing. The computing chip consists of three parts: a spatial diffraction optical neural network core particle, a plane matrix reconfigurable optical neural network component and a plane switchable optical switch module, the single or multiple spatial diffraction optical neural network core particles, the plane matrix reconfigurable optical neural network assembly and the plane switchable optical switch module are combined together in a three-dimensional heterogeneous integration mode, and efficient multi-task optical calculation processing is achieved. The spatial diffraction optical neural network and the plane matrix reconfigurable optical neural network are combined with each other, a traditional optical calculation framework is broken through, the method has wide application prospects in the optical calculation field, and the blank of related technical fields is filled.

The invention relates to a RAW image multi-task processing network obtaining method and device. The method comprises the following steps: inputting an RGB image into a mirror image backbone network to obtain an RGB feature map; an RAW image obtained through RGB image degradation is input into a backbone network of an image multi-task processing network to be trained to obtain an RAW feature map, and the backbone network and a mirror image backbone network form a twin network; splicing the RGB feature map and the RAW feature map to obtain a spliced feature map; inputting the spliced feature map into an automatic coding module to obtain a degradation prediction parameter; constructing a loss function including loss between the degradation prediction parameter and the degradation parameter; and training the image multi-task processing network to obtain an RAW image multi-task processing network. The RAW image multi-task processing network obtained by the invention can be directly applied to the RAW image obtained by the image acquisition equipment, and the image multi-task processing efficiency is improved.

The invention relates to an ultrasonic detection method, system and device applied to repaired mouth detection and a medium, and the method comprises the steps: obtaining first position information corresponding to a repaired mouth, and calculating a moving path of an ultrasonic detector corresponding to the first position information based on a preset first heuristic search algorithm; based on the moving path, acquiring a first ultrasonic image corresponding to the repaired mouth acquired by the ultrasonic detector; and joint coating identification information corresponding to the first ultrasonic image is obtained based on a trained identification model, and the joint coating identification information is abnormal judgment information or normal judgment information. According to the method, the ultrasonic detector can move according to the optimal moving path, the accuracy and efficiency of ultrasonic image recognition are improved, the standardization degree of repairing work of the repaired mouth and the repairing quality of the repaired mouth are improved, and efficient reinspection of the repaired mouth is achieved.

The invention discloses an analog domain time signal photon peak processing method and system architecture based on synaptic delay plasticity. The system architecture comprises an analog domain time signal input, a detector array, a photon neuron peak coding layer, a synaptic delay adjusting module, a photon neuron peak output layer, an optical decision module and a multifunctional output module. According to the method, the real-time weight-free photon spike processing of the simulation domain time signal can be realized by combining the advantages of biological nerve mimic synaptic parameter diversity and all-optical high-speed processing, combining the characteristics of photon neuron threshold judgment and time domain integration and a pulse neural network algorithm. The method is of great significance in improving real-time processing of simulation domain time signals such as flight time signals and promoting application of photon peak processing in low-delay scenes.

The invention discloses a mine time shift resistivity monitoring data screening method, a storage medium and software. The problems that in the prior art, data screening mainly depends on artificial experience, the intelligent degree is insufficient, and efficiency is not high are solved. The method comprises the following steps: firstly, performing dimension reduction on high-dimensional data to obtain essential features of the data, and removing data redundancy to realize feature extraction; feature selection is carried out in the feature space after dimension reduction; and finally, screening the monitoring data in the feature space by using a spectral clustering method. The method can be realized by means of a software technology.

The application discloses a multi-target detection method in a complex electromagnetic interference environment for a pulse Doppler radar, a joint reverse diffusion process is introduced to jointly update interference samples and amplitude samples, and thus the problems of target signals being covered or submerged by interference signals, rising of a detection false alarm rate, and difficulty in effectively separating interference from targets can be overcome; a denoising score matching criterion is introduced to train a neural network of an interference score function, so as to solve the problems of high interference intensity and complex structure in multi-target detection in a complex electromagnetic interference environment; and a sparse Bayesian learning method is introduced to adaptively model a sparse structure of multi-target echo signals. The application further provides a multi-target detection device in a complex electromagnetic interference environment. The method provided by the application can realize modeling and suppression of multiple types of interference, and simultaneously improve the detection probability of multiple low observable targets in the distance and Doppler dimension, the interference suppression capability, and the system robustness in a complex electromagnetic environment.

The invention discloses a feature extraction method for digital substation equipment. The method comprises the following steps: analyzing target digital substation equipment to confirm a fault diagnosis signal; obtaining a fault diagnosis signal and carrying out noise reduction and decomposition to obtain candidate intrinsic mode function components; calculating the sample entropy of the candidate intrinsic mode function component, and comparing the sample entropy with the sample entropy of the actual signal to obtain an intrinsic mode function component; and feature vectors in the intrinsic mode function components are extracted, dimensionality reduction, inspection and analysis are carried out, and the final features of the digital substation equipment are obtained. The invention also discloses a system for realizing the feature extraction method of the digital substation equipment. According to the method, noise reduction, decomposition, sample entropy calculation, comparison and dimension reduction analysis are carried out on the obtained transformer substation equipment signal, so that feature extraction of digital transformer substation equipment is realized, the reliability is higher, and the accuracy is better.

The invention discloses a feature extraction method of a neural network based on double-domain hybrid interaction, and the method is applied to the field of multi-spectral image target recognition. The method comprises the following steps: firstly, constructing a two-domain-based two-dimensional multi-scale convolution embedding module to realize feature representation on different scales and enhance the richness of extracted features; secondly, designing a mixed domain sharing module to realize information interaction, promoting information interaction between an image spectral domain and a spatial domain, reducing heterogeneity of the spatial domain and the spectral domain, and realizing improvement of network feature extraction capability; and finally, introducing a pyramid scale fusion block to directly integrate features of interaction between multi-channel, multi-scale, local and global spectral spaces, and realizing extraction of advanced semantic features. According to the method, the image double-domain information is interactively interpreted in the network extraction process, so that the extracted image feature information has better separability, and better robustness and effectiveness are shown in the field of image recognition.

The invention relates to the technical field of data analysis, and discloses a solid state disk remaining life analysis method based on an improved classifier model, which comprises the following steps: carrying out abnormal value screening and time sequence smooth interpolation operation on hard disk monitoring data to obtain a hard disk data sequence; performing data standardization operation on the hard disk data sequence to obtain a standard data sequence, and performing multi-channel feature extraction and mutual information dimension reduction operation on the standard data sequence to obtain a hard disk attribute feature sequence; performing bidirectional time sequence coding and global attention coding on the hard disk attribute feature sequence to obtain hard disk attribute time sequence features; variational fault classification, fault labeling and time sequence expansion convolution are carried out on the hard disk attribute time sequence features to obtain standard time sequence features, and life analysis is carried out on the standard time sequence features to obtain the remaining life of the hard disk. The invention further provides a solid state disk remaining life analysis system based on the improved classifier model. The method can improve the accuracy of life analysis.

The present invention discloses a real-time prediction method for wind power, comprising the following steps: step S1, for any wind power equipment that has been put into operation, pre-acquire the wind power and wind speed of the wind power equipment at each power generation moment after the power generation operation is completed; step S2, establish a prediction model for wind power, the prediction model for wind power is used to establish a nonlinear relationship between the wind power of the wind power equipment at a previous moment and the wind power at a later moment; step S3, collect the wind power and wind speed of the wind power equipment at the current moment in real time, and then input them into the prediction model for wind power, so as to predict and obtain the wind power of the wind power equipment at the later moment in real time. The real-time prediction method for wind power disclosed by the present invention is scientifically designed, and by establishing a prediction model for wind power, it is possible to efficiently and reliably predict the wind power of the wind power equipment in real time, and has great practical significance.

The present invention relates to the field of data analysis technology, and invents a method for analyzing the remaining life of a solid-state hard disk based on an improved classifier model, comprising: performing outlier screening and time series smoothing interpolation operations on hard disk monitoring data to obtain a hard disk data sequence; performing data standardization operations on the hard disk data sequence to obtain a standard data sequence, and performing multi-channel feature extraction and mutual information dimension reduction operations on the standard data sequence to obtain a hard disk attribute feature sequence; performing bidirectional time series encoding and global attention encoding on the hard disk attribute feature sequence to obtain hard disk attribute time series features; performing variational fault classification and fault labeling and time series dilation convolution on the hard disk attribute time series features to obtain standard time series features, and performing life analysis on the standard time series features to obtain the remaining life of the hard disk. The present invention also proposes a solid-state hard disk remaining life analysis system based on an improved classifier model. The present invention can improve the accuracy of life analysis.

The invention provides a power transmission line tower aeolian vibration video analysis method and an early warning device, and relates to the technical field of power grids. According to the method, the video image of the power transmission line tower is framed to obtain the multi-frame video image, the position information of each key part of the target tower in the video image is determined, the time-frequency analysis is carried out, the vibration characteristics such as the amplitude-frequency characteristic, the time-frequency characteristic and the vibration dynamic characteristic are determined, and the comprehensive characteristic extraction of each key part of the target tower is realized. And then vibration analysis results such as the vibration type and the vibration intensity are obtained through analysis in combination with the inherent characteristics of the target tower, and aeolian vibration video analysis of the transmission line tower is achieved. The process does not need manual inspection, and the inspection efficiency of the power transmission line tower is improved.

The invention discloses a metallogenic environment analysis method and system based on aviation hyperspectral remote sensing. The method comprises the steps that an existing stratigraphic map of a rift valley environment, an existing stratigraphic map of a sea ditch environment, an existing stratigraphic map of an island arc environment, an existing stratigraphic map of a mainland marginal sea environment, an existing stratigraphic map of a ground suture line type metallogenic environment and an existing stratigraphic map of a plate internal environment are collected and preprocessed, and then a metallogenic feature map is extracted; constructing a metallogenic characteristic stratigraphic analysis model based on a deep learning architecture, and training the metallogenic characteristic stratigraphic analysis model by using a metallogenic characteristic map; and using aviation hyperspectral remote sensing SASI data of a to-be-detected area and performing preprocessing, inputting the preprocessed data into the trained metallogenic characteristic stratum analysis model to obtain the metallogenic characteristics of the to-be-detected area, and analyzing the metallogenic environment according to the obtained metallogenic characteristics to obtain a metallogenic environment analysis result.

This invention discloses a method, device, and system for thymoma detection and classification based on a prediction model, relating to the field of thymoma detection and classification technology. It addresses the problem of the inability to perform detailed and accurate verification and classification of the prediction results of the prediction model. The method includes: analyzing the grayscale differences between different pixels in a medical image based on grayscale values; analyzing the positional relationship between lesion pixels and the image coordinate system based on the coordinates of different pixels; analyzing the undetermined thymoma region in the image to be identified based on the reference grayscale deviation interval and the included angle interval to obtain the true thymoma region and the pseudo-thymoma region; analyzing the adhesion type between the true thymoma region and the tissue region to determine whether the adhesion type of the true thymoma region is a non-adhesive thymoma region or an adherent thymoma region; and outputting the classification result of the image to be identified based on different adhesion types. This invention achieves detailed and accurate verification and classification of the prediction results corresponding to the prediction model.

The invention discloses a radar pulse stream panoramic cutting method, device and system and a storage medium. The method comprises the following steps: S1, quantifying a radar pulse stream; and S2, carrying out panoramic cutting on the quantized radar pulse flow based on a PatchPST network. By adopting the technical scheme of the invention, the radar pulse stream panoramic cutting can be effectively realized.

The invention provides a traffic sign recognition method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a traffic sign image, carrying out the global pooling operation of the traffic sign image, obtaining the global feature information, inputting the global feature information into a channel attention module, and obtaining a channel attention module; attention image features are obtained by calculating the attention weight of each channel in the global feature information, the size value of a filter in a channel attention module is adaptively adjusted based on the number of channels of the global feature information, and the attention image features are input into a big kernel attention module. The attention image features are deeply extracted through a plurality of convolutional layers and then spliced to obtain large-kernel image features, traffic sign features are obtained based on the large-kernel image features and the attention image features, and target detection is performed on the traffic sign features to obtain a traffic sign recognition result. Through the traffic sign recognition method, the technical problem of low accuracy of long-distance traffic sign recognition is solved.

The present invention discloses a method for extracting features from three-dimensional point cloud data, including obtaining a known three-dimensional point cloud data set and preprocessing it; setting source charges at the vertices of the minimum bounding box of the three-dimensional point cloud data set and constructing probe point charges for each point in the three-dimensional point cloud data set; calculating the point geometry Coulomb force of each point in the three-dimensional point cloud data set; merging the point geometry Coulomb force with the corresponding three-dimensional coordinates to obtain a training data set, and training a deep learning network to obtain a feature extraction network; and using the feature extraction network to complete feature extraction of target three-dimensional point cloud data. The present invention also discloses a classification and segmentation method including the feature extraction method for three-dimensional point cloud data, and a system for implementing the feature extraction method for three-dimensional point cloud data. The present invention has low-dimensional and explicit features, has stronger robustness to real data, and can not only realize feature extraction of three-dimensional point cloud data, but also has higher reliability, better accuracy, and better applicability.