153results about How to "Rich feature information" patented technology

The invention provides a traffic sign recognition method for a driverless car, and belongs to the technical field of image processing. According to the traffic sign recognition method, on the basis of algorithms such as a convex hull algorithm and a Hu invariant moment and transverse and longitudinal histogram scaling fast matching algorithm. Compared with existing traffic sign recognition methods, the traffic sign recognition method for the driverless car has the advantages of being large in recognition range, capable of recognizing a ban sign and an indicative sign, good in real-time performance, high in recognition accuracy and low in mistaken recognition rate.

The invention relates to a Chinese sign language recognition method based on kinect. The method comprises the following steps of acquiring depth information of hands and the 3D (three-dimensional) coordinate information of bone joint points of main parts of a human body through kinect; respectively acquiring hand characteristic information, position characteristic information and direction characteristic information by processing the characteristics of the acquired information; acquiring a hand element, a position element and a direction element for the acquired hand, position and direction characteristic information by adopting different polymerization algorithm, carrying out the element matching by utilizing a neighboring method and a template matching method adopting an Euclidean distance as a similarity measurement criterion, and determining the hand sign implication. Due to the adoption of the method, a purpose for effectively recognizing the Chinese sign language with vast vocabulary can be realized; and moreover, each element is independently acquired in parallel, so that the Chinese sign language can be recognized in real time. By implementing the method, different sign languages can be recognized in real time, so that Chinese deaf-mute can effectively communicate with other people.

The invention relates to a micro-expression recognition method based on a space-time appearance movement attention network, and the method comprises the following steps: carrying out the preprocessingof a micro-expression sample, and obtaining an original image sequence and an optical flow sequence with a fixed number of frames; constructing a space-time appearance motion network which comprisesa space-time appearance network STAN and a space-time motion network STMN, designing the STAN and the STMN by adopting a CNN-LSTM structure, learning spatial features of micro-expressions by using a CNN model, and learning time features of the micro-expressions by using an LSTM model; introducing hierarchical convolution attention mechanisms into CNN models of an STAN and an STMN, applying a multi-scale kernel space attention mechanism to a low-level network, applying a global double-pooling channel attention mechanism to a high-level network, and respectively obtaining an STAN network added with the attention mechanism and an STMN network added with the attention mechanism; inputting the original image sequence into the STAN network added with the attention mechanism to be trained, inputting the optical flow sequence into the STMN network added with the attention mechanism to be trained, integrating output results of the original image sequence and the optical flow sequence through the feature cascade SVM to achieve a micro-expression recognition task, and improving the accuracy of micro-expression recognition.

The invention discloses a traffic image semantic segmentation method based on a multi-feature map. The method comprises the following steps: obtaining a multi-feature map training sample: a disparitymap, a height map and an angle map; constructing a network model, training the network model, inputting the trained network model and a six-channel test image into the network model, outputting a probability value that each pixel belongs to each object category in the six-channel image via a multi-class classifier softmax layer, then predicting the object category to which each pixel in the six-channel image belongs, and finally outputting an image semantic segmentation map. By adoption of the traffic image semantic segmentation method based on the multi-feature map provided by the invention,the fusion of a color image with a depth map, the height map and the angle map, more feature information of the image can be obtained, and it is conducive to understanding the road traffic scene and improving the semantic segmentation accuracy. According to the traffic image semantic segmentation method based on the multi-feature map provided by the invention, by means of the learned effective features, the object category to which each pixel in the image belongs can be predicted, and the image semantic segmentation map is output.

The invention provides machine learning based human disease detection methods and a detection product. The method includes extracting the intrinsic data characteristics of electrocardial vector data,and quantitative index data thereof; constructing a machine learning classification model of the electrocardial vector data characteristics; and assigning corresponding weight values to different classification results identified by the model so that the comprehensive judgment results of human disease detection can be obtained. According to the provided detection method, the technical problems ofthe model processing of electrocardial continuous dynamic signals, the modeling analysis of quantitative data of data characteristics and the auxiliary diagnosis of human diseases can be solved. The detection methods and detection product can improve the accuracy and detection efficiency of human disease detection; and diagnosis effects can be enhanced with the increasing of the quantitative information of the electrocardial vector data characteristics expanded into a database.

The invention discloses a method for establishing a pulmonary nodule detection device on the basis of a 3D convolutional neural network. The method comprises the steps of establishing a training sample; establishing a pulmonary nodule detection network, wherein a pulmonary nodule segmentation network comprises a convolution unit, a 64*64*64(32) residual convolution unit A, a 32*32*32 (64) residualconvolution unit B, a 16*16*16 (64) residual convolution unit C, an 8*8*8 (64) residual convolution unit D and a 16*16*16 (64) residual convolution unit E which are connected with one another in sequence, output feature graphs of the residual convolution unit E and output feature graphs of the residual convolution unit C are spliced according to channels and then input to the residual convolutionunit F, and output feature graphs of the residual convolution unit F and output feature graphs of the residual convolution unit B are spliced according to channels and then input to an RPN network toachieve pulmonary nodule detection of the input graphs; training the pulmonary nodule detection network and obtaining the pulmonary nodule detection device.

The invention discloses a polarization SAR image classification method based on a complex contour wave convolution neural network, and a problem of low classification accuracy in the prior art is mainly solved. The method comprises the steps of (1) inputting and normalizing a polarization coherent matrix T of a polarization SAR image to be classified, (2) according to the normalized matrix, constructing characteristic matrixes of a training data set and a test data set, (3) constructing a complex convolution neural network, and thus obtaining a complex contour wave convolution neural network, (4) training the complex contour wave convolution neural network by using the training data set, and obtaining a trained model, and (5) inputting the characteristic matrix of a test data set into the trained model to carry out classification, and obtaining a classification result. According to the method, the convolution neural network is extended to a complex domain to carry out operation and extract image characteristics of multiple scales, multiple directions and multiple resolution characteristics, the classification precision of the polarization SAR image is effectively improved, and the method can be used for target detection and identification.

The invention discloses a human skeleton action recognition method, and relates to a method for identifying graphics. The method is a human skeleton action recognition method combining space-time attention and a graph convolution network. Diversity and complementarity of different feature information are fully mined; the weight value of each joint point of the space structure and the importance ofeach frame of the video sequence are adaptively adjusted by using an attention mechanism; the motion recognition of the human skeleton is carried out by using the graph convolution network, and the defects that in the prior art, space-time feature information cannot be better captured, and errors are likely to occur in the recognition of the difficult motion of the human body are overcome.

The invention discloses a pedestrian hybrid search method and system in a video monitoring scene, and belongs to the field of video content search. The method comprises the following steps: carrying out the video decoding to obtain a frame image; respectively carrying out face target detection and pedestrian target detection on the frame image to respectively obtain a face snapshot and a pedestrian snapshot of a pedestrian target; recognizing face snapshots and pedestrian snapshots belonging to the same pedestrian target in the same frame, and associating the face snapshots and the pedestriansnapshots together; extracting face features according to the face snapshots, extracting pedestrian re-recognition features according to the pedestrian snapshots, and matching the pedestrian targets in the current frame with the pedestrian targets in the processed frame images through feature matching so as to obtain tracks of the same pedestrian target; and storing the pedestrian target ID and the face snapshot, the pedestrian snapshot, the face feature, the pedestrian re-recognition feature and the track information of the pedestrian target into a database to obtain a search database. According to the method, multi-angle feature description of the pedestrian target can be provided, and reliable support is provided for application.

A method for detecting small objects in a structured road based on deep learning comprises the steps that image data, containing the small objects, on the real structured road are collected, and the positions and the category information of the small objects in the structured road are marked through a manual method; Constructing a deep convolutional neural network suitable for small object detection in the structured road and a corresponding loss function; Inputting the acquired image and the labeled data into the convolutional neural network constructed in the previous step, updating the parameter value in the neural network according to the loss value between the output value and the target value, and finally obtaining an ideal network parameter. The invention provides a brand new network structure for the problem that the current neural network is poor in small object detection. On the premise that the calculated amount is not increased basically, the performance of small object detection is greatly improved, and the method can be conveniently deployed in an existing intelligent driving system, so that an intelligent driving automobile can detect dangerous objects on a road in along distance and respond in time, and the safety in the driving process is improved.

The invention discloses a method for removing station captions and subtitles in an image based on a deep neural network, and relates to the technical field of image restoration, and the method comprises the following steps: S1, building an image restoration model; S2, preprocessing images of the training set; S3, processing training data: taking the training image as a real image Pt; Setting a pixel point RGB value in a Mask1 region in the training image as 0, and taking the pixel point RGB value as a training image P1; Setting a pixel point RGB value in a Mask2 region in the training image as0, and taking the pixel point RGB value as a training image P2; S4, training the image restoration model to obtain a trained image restoration model; S5, image restoration; The method comprises the following steps of: preprocessing an image or a video needing to remove station captions and subtitles; According to the image restoration method, based on the deep learning idea, station captions andsubtitles in the image are automatically and rapidly removed, the processing process is clear and clear, restoration real-time performance is high, and the application range is wide.

The invention provides a machine vision and millimeter-wave radar fused multi-vehicle target tracking method, which comprises the following steps of: acquiring road target information by using millimeter-wave radar, and screening vehicle targets according to a kinematics parameter related filtering model; detecting vehicles in front of the road by using the visual information, and performing multi-vehicle target tracking based on a detection result; projecting the vehicle target into the image by adopting a machine vision and millimeter wave radar fusion model, setting an association judgmentstrategy to associate the visual tracking target with the vehicle target, and correcting the position and the size of the visual tracking bounding box in the image based on distance information detected by the millimeter wave radar. The technical problem that in the prior art, when multiple vehicles in front are continuously tracked, after errors are accumulated due to the fact that the size of avisual tracking boundary frame is too large or too small, an effective target is lost can be solved.

The invention relates to the field of intelligent traffic management and particularly discloses a sound signal vehicle type identification system combined with an oscillation mark line. According to the sound signal vehicle type identification system combined with the oscillation mark line, a sound signal collecting module collects the sound signals generated when a vehicle passes by the oscillation mark line, and converts the sound signals which are analog signals into digital signals; a sound signal processing module judges whether the vehicle passing by the oscillation mark line exists or not, when the vehicle passes by the oscillation mark line, feature data of the current sound signal are extracted; a vehicle type identification module conducts vehicle type identification and classification based on the feature data; the oscillation mark line comprises N protrusion particle bands, wherein N is a positive integer which is larger than or equal to 2; protrusion particles of the ith protrusion particle band and protrusion particles of the i+1th protrusion particle band are arranged in a staggered mode, wherein i is a positive integer which is smaller than or equal to N-1. Oscillation mark lines are arranged on a highway and urban roads. Noise generated when the vehicles pass by the oscillation mark lines are collected, processed and analyzed, arrival of the vehicles is effectively judged, abundant feature information is provided for sound signal vehicle type identification, and the identification precision is improved.

The invention discloses a subject classification method fusing multiple human brain atlases based on a graph convolutional neural network. The human brain atlas is a data structure and represents interaction information between different brain regions in the human brain. The method performs classification prediction on a subject by identifying five human brain atlases of the subject, and belongs to the field of brain science research and deep learning research. The classification method comprises the following steps: acquiring and preprocessing human brain functional magnetic resonance time sequence signals; constructing five types of human brain atlases for each sample according to different functional connection strength calculation methods so as to obtain five data sets; constructing five graph convolutional neural network classifiers; carrying out training on the corresponding human brain atlas data sets separately, and therefore obtaining the binary classification capacity of thespecific human brain atlas; and integrating prediction results of the five graph convolutional neural network classifiers, and performing classification prediction on the subject, i.e., predicting which kind of person the subject belongs to.

The invention relates to an attack data acquisition method and device of a honeypot system. The method comprises the steps that real industrial control equipment, virtual industrial control equipmentand an upper computer in the honeypot system are operated; wherein the upper computer is used for sending a control instruction to the real industrial control equipment and the virtual industrial control equipment and reading state data; and the flow monitoring device monitors the communication flow information of the honeypot system, and analyzes and records attack behavior data in the communication flow information. By adopting the method, the simulation degree of the honeypot system can be improved, and an cracking of an attacker is prevented.

The invention discloses a knowledge base construction method for scientific and technological information analysis, belonging to the field of computer knowledge base construction. A CWATT-BiLSTM-LSTMdmodel is provided for entity extraction, and an RL-TreeLSTM model is used for entity relation extraction. An encoding-decoding mode is adopted for entity extraction, a BiLSTM (Bidirectional Long-Short Memory Network) is used for coding, LSTMd (Long-Short Memory Network) is used for decoding, and the embedding layer and decoding layer are improved. Then the model is used to extract entities fromthe corpus in the field of science and technology information. On the basis of deep reinforcement learning, the RL-TreeLSTM model is provided to extract the relationship between entities. The RL-TreeLSTM model is divided into two parts: a selector and a classifier. The selector selects effective sentences to the classifier in order to reduce the noise caused by the remote monitoring method. The classifier extracts the entity relation from the effective sentences, which improves the accuracy of relation extraction.

The invention provides a neural network smoke image classification method fusing dark channels. The method comprises the following steps: preparing two types of samples, namely a smoke image and a non-smoke image, normalizing the samples into the same size, carrying out dark channel processing on all sample images, and dividing an original image and a corresponding dark channel image into a training set, a verification set and a test set to serve as data input of subsequent network training; secondly, training the data set by using the designed dual-channel convolutional neural network, addinga residual block to the first channel network to improve the classification performance, and inputting an RGB original image to extract a generalization feature in the original image; the second channel adopts an improved AlexNet simplified network and inputs a dark channel image to extract detail features of smoke in the dark channel; the two channels are trained respectively, and finally carrying out feature fusion to generate a training model to classify the images. The result shows that the method effectively improves the accuracy of smoke image classification.

The invention discloses an electric appliance load type intelligent identification device, comprising an information collection module, an information processing module and a communication module. The electric appliance load type intelligent identification device simultaneously adopts a start-up current characteristic, a fundamental wave voltage current phase difference of the electric appliance load and a load current frequency spectrum characteristic as the electric appliance load identification characteristics, and the characteristic information is abundant; a combined classifier comprising a support vector machine classifier and a Bayes classifier is adopted to perform identification classification; the comprehensive identification is performed by giving consideration to features of the two classifiers and the identification accuracy is high; and the obtaining method for the fundamental wave voltage current phase difference, the start-up current characteristic and the load current spectrum characteristic is simple and reliable. The device provided by the invention can be applied to public occasions such as student dormitories, large-scale pedlars' markets and the like which require electric appliance load management, and can also be applied to other electric appliance management required occasions demanding electric appliance load type identification and statistics.

The invention relates to an image characteristic description method based on a Gabor synthetic characteristic, which comprises the following steps: a first step, acquiring and uploading a human face image signal; a second step, performing resolution adjustment and matrix expression of a face image; a third step, extracting an image characteristic; and a fourth step, synchronously outputting a processing result. According to the image characteristic description method, through using the amplitude part and the phase part which are converted by the Gabor wave filter, wherein the phase part comprises direction information in the Gabor filtering result, a certain characteristic discriminating meaning is realized; the filtering result of a Gabor wave filter set is sufficiently utilized; more abundant characteristic information is extracted for facilitating afterward identification; furthermore for aiming at a defect of averaging image blocks, different importance degrees of sub image blocks to the integral image are considered; and furthermore the face characteristic can be better described in combination with texture contribution degrees.

The invention discloses an electricity load type identification method, which is realized through an electricity load identification device consisting of an information collection module, an information processing module and a communication module. The electricity load type identification method simultaneously adopts electricity load starting current characteristics including starting process time, a starting current maximum value, and starting current maximum value time and a load current frequency spectrum characteristic of the electricity load as identification characteristics for the electricity load, and the characteristic information is rich. The electricity load type identification method adopts a combination classifier comprising a support vector machine classifier and a Bayes classifier to perform identification classification, performs comprehensive identification in consideration of characteristics of two classifiers, and thus has high identification accuracy. The provided methods for obtaining starting current characteristics and load current frequency spectrum characteristics are simple and reliable. The electricity load identification device can be used in some collective public places like a students 'dormitory, a large-scale pedlars' market, etc, where the electricity load management is needed, and can also be used in other places where need to perform electricity load type statistics and electricity appliance management.

The invention discloses a polarized SAR image classification method based deep multi-example learning, and solves the technical problem of low classification precision due to insufficient feature extraction in a conventional polarized SAR image classification method. The method includes the steps of filtering polarized SAR images, selecting a training sample set, extracting sample characteristics, initializing a convolutional neural network (CNN) and a deep belief network (DBN), normalizing sample characteristics, training the CNN and the BDN, extracting combined characteristics, inputting the combined characteristics to an SVM classifier for training, classifying polarized SAR images through the trained SVM classifier, outputting the classification result and calculating classification precision. The method effectively combines image spatial neighborhood characteristics and polarized characteristics, thereby improving the classification precision for polarized SAR images. The method is applicable to terrain classification and target identification for polarized SAR images.

The invention discloses a multispectral image classification method based on dual-channel deep convolutional generative adversarial network (DCGAN) and feature fusion. The method comprises steps of: inputting multispectral images; normalizing the image of each band of each multispectral image; obtaining a multispectral image matrix; obtaining a data set; creating a dual-channel DCGAN model; training a dual-channel DCGAN classification model; and classifying a test data set. The multispectral image classification method introduces the dual-channel DCGAN, combines the feature fusion, extracts avariety of multispectral high-level feature information in multiple directions, enhances the feature characterization ability, and improves the classification effect.

The invention relates to the technical field of tea quality monitoring, in particular to a method for monitoring the fermentation degree of black tea through a hyperspectral coupled nanocrystallization colorimetric sensor. Volatile substances in the fermentation process of black tea are captured by utilizing a nanocrystallization colorimetric sensing array, colorimetric array feature information is efficiently extracted by combining a hyperspectral image technology with dimension reduction algorithms such as principal component analysis and linear discriminant analysis, and an information fusion discrimination model with strong robustness and high accuracy is established by adopting algorithms such as partial least squares discrimination, multivariate linear discrimination, a support vector machine, an extreme learning machine, an artificial neural network and a deep belief network to realize rapid and accurate discrimination of the fermentation degree of black tea. The method has the characteristics that the analysis speed is high, the sensitivity is high, the cost is low, a sample does not need to be pretreated, and online nondestructive detection is facilitated.

The invention discloses a Chinese grammar debugging method and system based on multivariate text features. The method comprises the steps: (1) carrying out vector representation on a text through a pre-training model and grammar priori knowledge to obtain a semantic feature vector and a part-of-speech feature vector, and carrying out end-to-end splicing on the part-of-speech feature vector and thesemantic feature vector to obtain a vector sequence of the text; (2) extracting a feature vector sequence of the text by using a Bi-LSTM model; (3) carrying out attention enhancement based on semantic and part-of-speech matching information on the feature vector sequence; (4) performing linear transformation on the feature vector sequence after attention enhancement to obtain a label prediction sequence; (5) carrying out information enhancement based on word order relation characteristics on the label prediction sequence; and (6) capturing the constraint information of the label prediction sequence after information enhancement, and judging a grammar error boundary position based on the constraint information. Through verification, the method shows a good error checking effect and is superior to other existing similar methods.

The invention discloses a feature extraction method for ultrasound tissue characterization based on the Hilbert-Huang transform. The method comprises the steps of acquiring multiple frames of ultrasonic echo RF signals of a tissue sample, selecting regions of interest, and constructing an ultrasonic RF time sequence; extracting features of the ultrasound RF time sequence by using a HHT algorithm to obtain sample feature vectors based on the relation of time, frequency and energy; and performing feature screening and feature fusion by using a rankfeatures algorithm, calculating a sample feature fusion index, and taking the feature fusion index as the feature for tissue characterization. The feature fusion index provided by the invention can effectively reflect the difference between different types of tissue samples and is suitable for the feature extraction of the ultrasonic tissue characterization.

A small sample industrial product defect classification method based on two-stage transfer learning comprises the following steps: S1, collecting positive and negative samples to form a data set; S21,expanding the number of the negative samples in the data set by 2-3 times by using an image data enhancement means, randomly selecting positive samples of which the number is equivalent to the numberof the expanded negative samples, and forming a data subset of which the number is balanced; S22, forming another data set subset by using the remaining positive samples; S31, selecting a CNN detection model, and carrying out first-stage training; S32, carrying out training in the second stage on the data set subset containing the remaining positive samples and the expanded negative samples; andS4, after the model training in the step S32 is converged, testing the classification performance of the model on the test set, if the requirements are met, performing online test, otherwise, repeatedly dividing the data subsets and the model training process, and repeating the steps S21 to S32 until the requirements are met. The method has the following beneficial effects: 1, the method has defect image high-dimensional features with better performance; 2, the representation capability of the model on an industrial product image is improved; and 3, the model training strategy has good universality.

The invention discloses an action recognition method based on double-flow convolution attention. The method comprises the following steps: firstly, preprocessing a video to obtain a frame image sequence and an optical flow image sequence, and respectively extracting appearance feature representation and action feature representation of the video; then constructing a convolution attention module to obtain attention feature representations of the frame image and the optical flow image, and performing information fusion on the two attention representations through a double-flow fusion module; and then training an action recognition model using a convolution attention mechanism and a double-flow fusion method, and outputting the action category of the preprocessed new video according to the model. According to the method, channel attention and space-time attention are utilized to capture a potential mode and a space-time relationship of video action contents, and information fusion is carried out on appearance features and action features of the video from a global perspective through double-flow fusion, so that the problem of time sequence information loss of long-term time sequence dependence of the video is effectively relieved; and the accuracy of action recognition is improved.

The invention relates to an image segmentation method, computer equipment and a readable storage medium. The method comprises the steps of obtaining a to-be-segmented medical image; and inputting themedical image into a segmentation model to obtain a segmentation result of the medical image. An up-sampling module of the segmentation model comprises a mixed channel disruption mechanism layer. In this method, the up-sampling module of the segmentation model comprises the mixed channel disruption mechanism layer, so that more feature information of the input medical image can be obtained while the calculated amount and the parameter amount of the segmentation model are reduced, the input medical image can be segmented more accurately, and the accuracy of the segmentation result of the obtained medical image is improved.