372 results about "Neural network classifier" patented technology

An artificial neural network device that utilizes one or more non-volatile memory arrays as the synapses. The synapses are configured to receive inputs and to generate therefrom outputs. Neurons are configured to receive the outputs. The synapses include a plurality of memory cells, wherein each of the memory cells includes spaced apart source and drain regions formed in a semiconductor substrate with a channel region extending there between, a floating gate disposed over and insulated from a first portion of the channel region and a non-floating gate disposed over and insulated from a second portion of the channel region. Each of the plurality of memory cells is configured to store a weight value corresponding to a number of electrons on the floating gate. The plurality of memory cells are configured to multiply the inputs by the stored weight values to generate the outputs.

A neural network classifier provides the ability to separate and categorize multiple arbitrary and previously unknown audio sources down-mixed to a single monophonic audio signal. This is accomplished by breaking the monophonic audio signal into baseline frames (possibly overlapping), windowing the frames, extracting a number of descriptive features in each frame, and employing a pre-trained nonlinear neural network as a classifier. Each neural network output manifests the presence of a pre-determined type of audio source in each baseline frame of the monophonic audio signal. The neural network classifier is well suited to address widely changing parameters of the signal and sources, time and frequency domain overlapping of the sources, and reverberation and occlusions in real-life signals. The classifier outputs can be used as a front-end to create multiple audio channels for a source separation algorithm (e.g., ICA) or as parameters in a post-processing algorithm (e.g. categorize music, track sources, generate audio indexes for the purposes of navigation, re-mixing, security and surveillance, telephone and wireless communications, and teleconferencing).

A character recognition system is described, in particular a system suitable for use in monitoring cargo container codes or vehicle number plates. An image of the code is first analyzed to extract potential characters. As part of this process, long horizontal and vertical line segments are filtered out. The extracted potential characters are then input to a two-level character recognition means. The first level comprises a neural network classifier that classifies a character into a smaller set of possible characters; and then the second level comprises another neural network classifier which identifies which character among the smaller set of possible characters the extracted character is.

The invention discloses a method for quickly identifying and distinguishing the variety of corn, which belongs to the technical field of the detection of agricultural products. The method identifies the variety of the corn by an image processing technology quickly. Seed image information is obtained by a scanner, corn seeds are identified by a corn variety identification system on a computer, original image information is read, the image information of each pixel in each corn is stored as a 24-bit bmp format file, and the variety of the corn is distinguished by a trained neural network classifier after image processing and feature extraction are completed. The variety distinguishing rate of the model to an unknown sample is greater than 90%, which shows that a condition that the variety of the corn seeds is distinguished by image feature information is feasible. The invention has the characteristics of high automatization degree and quick, simple and convenient operation.

The invention discloses a driving scene classification method based on a convolution neural network, and the method comprises the following steps: collecting a road environment video image; carrying out the classification of a traffic scene, and building a traffic scene recognition database; extracting sample images of different driving scenes from the traffic scene recognition database, carryingout the feature extraction and multiple convolution training of the sample images through a deep convolution neural network, carrying out the rasterization of pixels, connecting the pixels to form a vector, inputting the vector into a conventional neural network, obtaining convolution neural network output, and achieving the deep learning of different driving scenes; carrying out the parameter optimization of a network structure of the built convolution neural network, obtaining a trained convolution neural network classifier, carrying out the adjustment of a traffic scene recognition model, and selecting an optimal mode as the standard of the traffic scene recognition model; carrying out the real-time collection of the image of a to-be-detected traffic scene, and inputting the image intothe traffic scene recognition model for the recognition of a road environment scene.

A forwarding looking ground penetrating mine detection apparatus includes a radiation source for irradiating a sample of ground suspected of containing at least one mine with a plurality of frequency swept ground penetrating radar signals. A detector receives target signals backscattered from the ground responsive to the radar signal. The detector includes a time-frequency analyzer which transforms the target signals into a time-frequency image representation (TFR). In a preferred embodiment, the detector can include a wavelet packet transformer (WPT) for extracting time-frequency localized information from the TFR in the form of feature set constructed from a wavelet table. The apparatus can also include a data dimensionality reducer for selecting features to form a feature subset from the feature set, preferably based on reference to a training data set. A multilayer neural network classifier can be based on the feature subset, and be adaptable to the surrounding environment through learning.

The invention provides a cutting-tool wear state evaluating method based on image analysis of a workpiece machining surface. The method comprises the following steps: firstly acquiring a texture imageof the workpiece machining surface by utilizing optical system imaging; then carrying out multi-feature extractions on the machining surface image and forming a BP neural network classifier facing the machining surface image according to the number of the multi-feature extractions; and adopting a fusion method based on a voting system to fuse the multi-feature classifier so as to realize the qualitative and the quantitative identification of cutting-tool wear and breakage states. The invention can better overcome the defect that the on-line image acquisition of cutting points of a cutting-tool cannot be carried out in a cutting machining process, accurately identify the wear and breakage states of the cutting-tool and can be applied to the cutting-tool on-line monitoring of numeral control machining tools.

The invention provides an image identification method based on depth learning and transfer learning. The image identification method based on depth learning and transfer learning includes the following steps: 1) a preparation step: reading a pre-training model, reading an image directory, and dividing a training set, a verification set and a test set; 2) a training step: constructing a full connection neural network classifier, taking the image set as input of the pre-training model, and using the output of the pre-training model to update the full connection neural network classifier; and 3) a storage step: storing the model result. The image identification method based on depth learning and transfer learning has the advantages of integrating with the application of depth learning and transfer learning so as to provide a relatively accurate bladder cancer diagnosis result for a user, on the basis of the extremely limited training time and training sample quantity.

Disclosed herein is an apparatus and method for detecting a person from an input video image with high reliability by using gradient-based feature vectors and a neural network. The human detection apparatus includes an image preprocessing unit for modeling a background image from an input image. A moving object area setting unit sets a moving object area in which motion is present by obtaining a difference between the input image and the background image. A human region detection unit extracts gradient-based feature vectors for a whole body and an upper body from the moving object area, and detects a human region in which a person is present by using the gradient-based feature vectors for the whole body and the upper body as input of a neural network classifier. A decision unit decides whether an object in the detected human region is a person or a non-person.

The invention relates to the field of double dynamic license plate identification, and provides a character confidence degree-based secondary license plate identification method and apparatus for the problems existent in the prior art. The method comprises the steps of performing character identification through template matching; giving out confidence degrees of identification results; for the results with relatively low confidence degrees, performing video super-resolution processing to obtain a frame of high-quality image; and based on the image, performing secondary license plate identification through a neural network classifier. According to the method and the apparatus, a license plate confidence degree threshold Th is preset; a picture is captured from a front end and license plate locating and segmenting are performed; characters of a license plate are identified; character identification confidence degrees and license plate identification confidence degree are calculated; when the character confidence degrees are all higher than the threshold Th, a license plate identification result is directly given, otherwise, the video super-resolution processing is performed; a frame of high-quality image is obtained by utilizing time domain information; and based on the image, the to-be-identified characters are input to classifiers for performing identification according to a position relationship of the to-be-identified characters, so that a final license plate identification result is obtained.

The invention discloses a method for recognizing an image of a plant leaf. The method comprises a training stage and a recognition stage. The training stage comprises the following steps of: inputting an image training set of the plant leaf; and storing a training result after performing batch preprocessing, batch feature extraction and neural network classifier training by using a system. The recognition stage comprises the following steps of: allowing a user to input a single image of the plant leaf of which the variety is unknown to the system and mark the leaf base point coordinate and leaf apex point coordinate of the leaf; performing preprocessing and feature extraction on the single image by using the system; and classifying by using the neural network classifier, outputting a plant variety list and attaching detailed graphics and text information to each item of the list. The method has the advantages of outputting a plurality of high-accuracy candidate results, greatly enhancing recognition credibility, supporting a scalar quantity of leaf features as well as leaf features in complex forms (such as a matrix form) and increasing the recognition rate of the plant leaf.

The invention relates to an abnormal intrusion detection ensemble learning method based on the Wiener process. The method comprises the following steps: selecting a network traffic data set; inputting each network traffic sample and sample probability distribution thereof to an uninitialized neural network classifier or a neural network weak classifier obtained through the previous training, judging whether the neural network weak classifier wrongly classifies each network traffic sample, and adjusting quantity and sample probability distribution of each network traffic sample; repeating the step 2 to obtain a plurality of neural network weak classifiers; determining the weight of each neural network weak classifier respectively; obtaining strong classifiers based on each weak classifier and the corresponding weight of each neural network weak classifier; inputting network data flow to be detected to the strong classifiers to obtain intrusion detection results; and repeating the step 6 until all the network data flow to be detected is detected. According to the method and apparatus in the invention, the problem of classification of the unbalanced data set can be solved, and an unbiased classifier with high classification correct rate can be obtained.

The present application relates to an intelligent outpatient guiding method and device, a computer device and a storage medium. The method includes: acquiring patient inquiry data; extracting inquiryelement features from the patient inquiry data; converting the inquiry element features into feature vectors; inputting the feature vectors into a preset neural network classifier to obtain a first department matching probability; inputting the feature vectors into a preset Bayesian classifier to obtain a second department matching probability; and generating a department recommendation result isaccording to the first department matching probability and the second department matching probability. The method can improve the accuracy of patient registration.

The invention relates to a gesture recognition method based on a BP neural network. The method includes the steps that electromyographic signals generated by multiple gesture actions of multiple samples are collected; the electromyographic signals are subjected to normalization; multiple feature values are extracted from the normalized electromyographic signals to form a feature value matrix; model training is conducted on the feature value matrix through the BP neural network algorithm to form a BP neural network model; the BP neural network model is saved in an electromyographic signal collection device for gesture recognition. During signal analysis and mode recognition, the improved normalization method effectively eliminates difference between electromyographic signals of different people, mode recognition is conducted through a BP neural network classifier, the gesture recognition rate is increased, and meanwhile the error probability in recognizing is greatly reduced.

The invention provides a crop disease identification method based on incremental learning. When new data arrive, continuous learning is carried out based on an original learning result, and the capability of progressive learning is achieved, which means that new knowledge can be obtained from new samples obtained by batch and the performance is gradually improved under a condition that original knowledge is effectively kept. Firstly, a crop disease sample database is collected, and simulation incremental learning of disease images in the sample database is carried out using a negative correlation integrated neural network as main technical means, so that an initial parameter of a negative correlation learning system is determined, an integrated neural network classifier based on negative correlation learning is initialized based on the initial parameter, and the classifier is trained using a sample in an initial stage; in an incremental learning stage, when an expert adds a new sample in the sample database, the integrated neural network classifier based on negative correlation learning only is updated by only training the newly-added sample data, so that the object of incremental learning is achieved; and finally, a diagnosis result of a disease picture and control measures are fed back to a user, so that the pest and disease can be accurately identified and diagnosed, and the object of comprehensive crop control is achieved.

The invention is applicable to the technical field of computers and provides an iris positioning method and device. The method comprises the steps of obtaining an iris image and calibrating the iris image; generating a binary mask layer label; generating training data according to the iris image and the binary mask layer label; establishing a deconvolution neural network; inputting the training data into the deconvolution neural network and calculating feature data of the training data; training a deconvolution neural network classifier according to the feature data and the binary mask layer label; and conducting iris positioning on the iris image through classifying the feature data by means of the deconvolution neural network classifier. The invention discloses an iris positioning technical scheme based on the deconvolution neural network and can realize supervised learning of iris positioning. The requirement on iris image quality is low, the size of the deconvolution neural network is small, and the number of training parameters and the computation amount are small, so that time and storage costs are reduced. Local extrema are avoided, and thus more precise iris positioning is achieved.

The embodiment of the invention discloses a method and device for labeling a semantic role. The method comprises the steps that at least one classification feature of a participle in an object statement to be labeled is acquired; semantic representation information of the acquired classification features is determined; semantic representation of all the classification features is adopted as input of a pre-generated neural network classifier, and semantic role labeling is carried out on the participle through the neural network classifier. According to the technical scheme, the complex and sparse feature based on a plurality of words, a plurality of word characteristics, a plurality of depending arc signs and a plurality of depending paths can be easily mapped into a dense feature, therefore, the dimension of feature space and feature establishment complexity are lowered, and a plurality of features can be combined automatically.

This patent concerns novel technology for detecting backdoors of neural network, particularly deep neural network (DNN), classifiers. The backdoors are planted by suitably poisoning the training dataset, i.e., a data-poisoning attack. Once added to input samples from a source class (or source classes), the backdoor pattern causes the decision of the neural network to change to a target class. The backdoors under consideration are small in norm so as to be imperceptible to a human, but this does not limit their location, support or manner of incorporation. There may not be components (edges, nodes) of the DNN which are dedicated to achieving the backdoor function. Moreover, the training dataset used to learn the classifier may not be available. In one embodiment of the present invention which addresses such challenges, if the classifier is poisoned then the backdoor pattern is determined through a feasible optimization process, followed by an inference process, so that both the backdoor pattern itself and the associated source class(es) and target class are determined based only on the classifier parameters and a set of clean (unpoisoned attacked) samples from the different classes (none of which may be training samples).

The invention provides a scene recognition method based on a single-hidden-layer neural network, and the method is characterized in that the method comprises a training stage and a recognition stage; the training stage comprises the steps: carrying out the preprocessing of a pre-collected sampling image set for training, extracting the local gradient statistical characteristics of the pre-collected sampling image set after preprocessing, enabling the local gradient statistical characteristics and a corresponding scene type label to be added to a single-hidden-layer neural network classifier for layered supervised learning, obtaining a plurality of different optimal parameters of various types of single-hidden-layer neural networks, and constructing a multilayer scene classifier according to the optimal parameters; the recognition stage comprises the steps: carrying out the preprocessing of a to-be-recognized image set, carrying out the local gradient statistical characteristics of the to-be-recognized image set after preprocessing, enabling a local gradient statistical characteristic vector to be inputted into the multilayer scene classifier for recognition, and obtaining a class mark of the scene. The method achieves the high-precision scene recognition.

The invention relates to an instruction-set-irrelevant binary code similarity detection method based on a neural network. The method mainly includes the steps that binary files are reversely analyzed, and 24 features on the 9 aspects of call relation features, character string features, stack space features, code scale features, path sequence features, path basic features, degree sequence features, degree basic features and map scale features of functions are extracted. Based on expression forms of the features, the similarity degrees of the 24 features of the two to-be-compared functions are calculated through 3 similarity calculation methods and serve as input vectors of an integrated neural network classifier, and predicted values of the overall similarity between the two functions are acquired and ranked. Compared with the prior art, dependence on specific instruction sets is avoided, similarity detection of binary files of different instruction sets can be achieved, accuracy is high, the technology is simple, and popularization is easy.