54 results about "Contextual image classification" patented technology

The invention relates to a pest image classification method based on a grading-prediction convolutional neural network. Compared to the prior art, by using the method of the invention, a defect that a pest image classification correct rate is low is overcome. The method comprises the following steps of collecting and preprocessing training images; marking image sample data; training a classification model based on the grading-prediction convolutional neural network; preprocessing images to be tested; and based on a grading model, automatically carrying out pest image classification. In the invention, a grading prediction framework is adopted; a segmentation result of the images is predicted firstly and then an integral image is combined; and final classification prediction is performed.

The invention discloses a bottom material sonar image classification method based on a convolution neural network, belonging to the technical field of image classification. The method comprises obtaining the sonar image of seafloor bottom, preprocessing the image with denoising and enhancement, extracting the edge shape based on a canny algorithm, and generating a gray scale-basic element co-occurrence matrix, constructing a convolution neural network classifier structure and a sample set, training a neural network, obtaining classification model and realizing bottom material sonar image classification. The present invention is directed to the graphical characteristics of bottom material sonar images of seafloor, the disadvantage of using single method is solved, and the learning strategyof the convolution neural network classifier is used to learn and train different types of seabed sediment, and finally the classification model with classification function is obtained, and the purpose of fast and accurate classification of seabed sediment sonar images is achieved.

The invention discloses a neural network-based ray image classification method and relates to the technical field of ray image classification. The method comprises the steps of (1) obtaining training samples of ray images and building a convolutional neural network model; (2) adjusting parameters of the convolutional neural network model in the step (1); (3) performing graphics preprocessing on the training samples in the step (1), inputting the preprocessed training sample to the adjusted convolutional neural network model in the step (2) for performing training, and obtaining feature information corresponding to the training samples; (4) extracting priori feature information corresponding to the training samples in the step (1); and (5) performing full connection on the priori feature information in the step (4) and feature information corresponding to a full connection layer of the convolutional neural network model in the step (3), and after class identifiers of the training samples corresponding to the priori feature information are stored, generating a ray image classification model.

The invention provides a medical image classification and segmentation method and a medical image classification and segmentation device. According to the invention, the problem in the prior art thatdue to the limited number of training data and the incomplete expression of the critical characteristics of medical images, the image classification and segmentation effect is poor can be solved. Themethod comprises the steps of carrying out data enhancement and pretreatment on a medical image, and constructing a classification data set and a segmentation data set; constructing a convolutional neural network model, extracting the characteristics of a natural image data set, and obtaining the activated characteristics of the convolutional neural network; based on the classification data set, training the convolutional neural network to extract the characteristics of the classification data set, so as to distinguish whether the medical image is a cancer image or a cancer-free image; and training the convolution neural network to extract the characteristics of the segmentation data set on the basis of the segmentation data set, so as to divide the canceration region of the medical image.

The invention discloses a deep neural network-based SAR (Synthetic Aperture Radar) texture image classification method, and aims to mainly solve the problem of low accuracy of SAR texture image classification with a larger number of samples and more characteristic dimensions in the prior art. The method is implemented by the following steps: (1) extracting low-level characteristics of an SAR image; (2) training the low-level characteristics of the SAR image to obtain advanced characteristics of the image by virtue of a first layer of RBF (Radial Basis Function) neural network of a deep neural network; (3) training the advanced characteristics to obtain more advanced characteristics of the image by virtue of a second layer of RBM (Restricted Boltzmann Machine) neural network of the deep neural network; (4) training the more advanced characteristics to obtain image texture classification characteristics by virtue of a third layer of RBF neural network of the deep neural network; (5) comparing texture classification characteristics of an image test sample with a test sample tag, and regulating parameters of each layer of the deep neural network to obtain the optimal test classification accuracy. The method is high in classification accuracy, and can be used for target identification or target tracking.

The invention discloses a classification method based on high resolution remote sensing image area information and a convolution neural network. Targeting problems that a quantity of image collection windows is too great and thus the classification efficiency is low when the convolution network is used for remote image classification using a fixed-size window to perform traversing. The invention provides a sampling window determination method based on image area characteristics and improves classification efficiency/ The classification method based on high resolution remote sensing image area information and the convolution neural network comprises steps of performing over-segmentation on an image to obtain area information of the image, determining a sampling window according to a certain criterion and sending sampling window data into the convolution neural network to perform classification, wherein a classification result is a classification result of a corresponding area. Targeting the restriction of the prior art in the high resolution remote sensing image, the classification method provided by the invention introduces a convolution neural network model in deep learning to extract image characteristics, provides a new technical scheme to the remote sensing image classification and improves classification accuracy and efficiency.

The invention discloses a neural image classification method, a computer terminal and a computer readable storage medium. The method comprises the steps of generating a connection matrix correspondingto a neural image according to a connection relationship between brain regions in the neural image; extracting a predetermined number of element values from the connection matrix according to a predetermined extraction rule and forming a tested feature vector; taking the plurality of tested feature vectors as nodes, and constructing a feature map according to similar characteristics among the non-image information corresponding to each node; initializing a pre-established classification model according to the feature map, and training the initialized classification model according to the tested feature vector; and classifying the neural images according to the trained classification model. According to the technical scheme provided by the invention, the feature map is constructed throughthe similarity between the non-image information of each tested feature vector, and the established classification model is initialized according to the feature map, so that the individual differenceof the tested object and the influence of acquisition equipment on the classification result are effectively solved, and the classification performance is improved.

Computerized systems and methods are disclosed for performing contextual classification of objects using supervised and unsupervised training. In accordance with one implementation, content reviewers may review training objects and submit supervised training data for preprocessing and analysis. The supervised training data may be preprocessed to identify key terms and phrases, such as by stemming, tokenization, or n-gram analysis, and form vectorized objects. The vectorized objects may be used to train one or more models for subsequent classification of objects. In certain implementations, preprocessing or training, among other steps, may be performed in parallel over multiple machines to improve efficiency. The disclosed systems and methods may be used in a wide variety of applications, such as article classification and content moderation.

The invention belongs to the technical field of medical information, and particularly relates to a medical image multi-stage classification method based on a symmetry theory. The method comprises the following steps that: images to be classified provide classification requests, wherein the images to be classified need to be original medical image data; and an image preprocessing process comprises the steps of image-based modeling, multi-stage classification and result display. The concepts of weak symmetry and high symmetry provided by the invention belong to redefinition on the medical images. A weak symmetry judging algorithm and a high symmetry judging algorithm are provided for realizing the multi-stage classification of the medical images. The classification accuracy of the multi-stage classification is very high, and the direct connection and layer-by-layer deep classification is realized in each stage, so the diagnosis precision of a doctor is improved, and the diagnosis time of the doctor is shortened. The medical image classification is realized by adopting the symmetry theory, so that the medical image multi-stage classification method based on the symmetry theory has higher accuracy.

A method of classifying an input image includes the initial steps of labeling an input image in accordance with a class and extracting at least one connected component from the input image. The method also includes the steps of calculating at least one feature of the input image and generating a model based on the at least one calculated feature. The method also includes the steps of repeating at least one of the previous steps for at least one other input image and comparing the at least one other input image with the model. The at least one other input image is classified in accordance with the class of the model if the at least one calculated feature of the at least one other input image is substantially similar to that of the model.

The invention discloses an artistic image classification method based on a convolutional neural network. The artistic image classification method comprises the following steps: obtaining each type ofartistic image, cutting the image, and constructing a training set and a verification set; then constructing a convolutional neural network by using hole convolution, a DKSE module and depth separableconvolution; and finally, inputting the training set into the built convolutional neural network to train the convolutional neural network to obtain a network model capable of being used for art image classification. Compared with a traditional method, the artistic image classification method has the advantages that the extracted overall features and local detail features of the art images are more comprehensive; the artistic image classification method can be suitable for classification research of various art images; the model universality is high, and the defects of existing classificationretrieval research of various art images are overcome; and the convolutional neural network designed in the invention can reduce professional requirements for art image classification personnel, andavoids complex feature extraction and data annotation work in a traditional classification algorithm; compared with other convolutional neural network structures, and the convolutional neural networkstructure is simple, and the classification accuracy is high.

The invention discloses an image classification method based on a local image block descriptor and a Fischer vector. The image classification method comprises the steps that a feature descriptor based on a local image block is constructed; modeling of the feature descriptor of a training data set is performed by using a Gaussian mixture model; a training image set and the Fischer vector for testing the image set are generated; a Gaussian space pyramid is constructed and image multi-scale space information is acquired; the training data set and a feature set for testing the data set are calculated; feature selection is performed on the feature set of the data set by using a mutual information method; and a classifier is trained so as to realize image classification. The image information can be accurately acquired so that the accuracy of image classification can be enhanced and the image classification method can be used for large-scale image classification and construction of the retrieval system.

A content evaluation system is described herein that empowers end users and organizations to share their interpretation of an automatically generated sentiment score. The system provides a control that a user can move to indicate agreement or disagreement with an automatic score. The system adds metadata to a revised score based on the user's feedback that tracks information about the user to consider different demographic contexts. The system performs rescoring with the user-provided scores with contextual consideration, and then exposes the rescored values on context specific endpoints. The system provides a crowd-sourcing approach that scales extremely well, adds more accuracy because individuals within known demographic categories/contexts do the scoring, and generates value-added data products that can be sold/re-sold.

The invention discloses a spatial Fisher vector based image classification method which mainly solves the problems of 'foreground and background visual ambiguities' of images and lack of feature point space distribution information in image description vectors in the prior art. The method includes the implementation steps: (1) extracting 'scale-invariant feature transformation' feature points of all images; (2) in a feature point space of images in a training set, clustering the feature points by the aid of a Gaussian hybrid clustering algorithm to obtain a code book; (3) utilizing gradient vectors and Cholesky components for generating a Fisher vector of each image; (4) performing 2X2 space region division on each image, and making statistics on number of feature points and coordinates of each cell block; (5) utilizing a Fisher vector of each cell block for stitching to generate a space Fisher vector of each image; and (6) utilizing a support vector mechanism for constructing a classification hyperplane to achieve image classification. The method has the advantages that image information can be described accurately, accuracy rate of image classification is improved, and the method can be used for large-scale image classification and retrieval system construction.

The invention discloses a spatial local clustering description vector based image classification method which mainly solves the problem of lack of feature point space distribution information in image description vectors in the prior art. The method includes the steps: (1) extracting 'scale-invariant feature transformation' feature points of all images; (2) in a feature point space of images in a training set, clustering the obtained feature points by the aid of a mean clustering algorithm to obtain a code book; (3) utilizing difference vectors for generating a local clustering description vector of each image in an image set; (4) performing 2X2 space region division on each image, and making statistics on number of feature points and coordinates of each cell block; (5) utilizing the local clustering description vector of each cell block for stitching to generate a spatial local clustering description vector of each image; and (6) utilizing a support vector mechanism for constructing a classification hyperplane to achieve image classification. The method has the advantages that image information can be described accurately, accuracy rate of image classification is improved, and the method can be used for large-scale image classification and retrieval system construction.

The invention discloses an SAR image classification method based on united sparse representation. The SAR image classification effect is improved based on an existing sparse representation method. According to the implementation process, the SAR image classification method comprises the steps that (1) SAR images to be trained are input, the features of the SAR images are extracted, and similar sets are classified; (2) united sparse representation is conducted on the similar set of each class of the SAR images, and a small dictionary and sparse coefficients of each similar set are obtained correspondingly; (4) SAR images to be tested are input, the features of the SAR images to be tested are extracted, feature vectors are projected on the small dictionary, and coefficients of the tested images are obtained; (5) the coefficients of the tested images and the sparse coefficients of all the trained images are matched, a set of most matched coefficients in the sparse coefficients are found, and the marked category of the set of most matched coefficients serves as the category of the SAR images to be tested. Compared with a traditional KNN and classic sparse representation classification method, the accuracy of even texture image and SAR image classification is greatly improved.

The invention discloses a gesture classification method based on transfer learning. The gesture classification method is suitable for gesture image classification. The method comprises the steps of 1,converting a gesture video V into a gesture frame data set G0; 2, performing noise removal, binarization and background segmentation processing on the G0 through a Gaussian filtering method, an OTSUalgorithm and image AND operation to obtain a gesture frame data set G1, and setting a label for the G1 to obtain a frame label data set L; 3, carrying out transfer learning by using a MobileNet convolutional neural network architecture and the weight file, and creating and training a model M1; 4, extracting features of the frame data set G1 through the model M1 to obtain a frame feature vector set F0; and 5, classifying the test set by taking the XGBoost as a classification model to obtain a final classification result. According to the method, the weight of the trained MobileNet convolutional neural network is migrated to a gesture image data set for feature extraction, and XGBoost is adopted as a classification model, so that the model calculation amount is reduced while the classification accuracy is improved.

The invention discloses an image classification method based on a topic model with a monitoring shared assembly. The method mainly solves the problems that in the prior art, the number of parameters is large, correlation of topics is ignored and representation effects of potential semantic features on images are poor. The method includes the implementation steps of establishing a natural image training set and a natural image test set, generating a vision dictionary, generating image sparse representation vectors, generating topic distribution vectors, establishing a natural image classification model and carrying out natural image classification. A Gibbs sampling method and a topic unbalance prior probability method are adopted in the image classification method, the number of the parameters is reduced, correlation of the topics is increased, representation effects of the topic distribution vectors on the images are better and image classification accuracy is improved.

The present invention relates to a model construction method for cell image classification. Aiming at the limitations of existing urine sediment recognition technology in real use scenarios, especially the problems that the accuracy is low for multiple classes and the identification speed cannot meet actual production needs, an efficient classification algorithm for urine formed element cells based on multi-model hybrid recognition is proposed. In the present invention, an efficient classification algorithm for urine formed cells based on multi-model hybrid recognition and software implementation include four parts: data preprocessing, obtaining of training model, obtaining of recognition results after threshold control judgment, and image recognition and acceleration .

The invention discloses a scene image classification method based on an interval Type-2 fuzzy support vector machine, and relates to the technical field of general image data processing or producing methods. According to the method, a novel fuzzy support vector machine design method based on an interval Type-2 fuzzy set is put forward, and a scene image is effectively classified based on three steps, namely, image classification feature extraction, feature dimension reduction and feature classification. According to the invention, uncertain information is described and processed based on an interval Type-2 fuzzy set, an interval Type-2 fuzzy support vector machine classification model is built, and the accuracy and robustness of scene image classification are improved.

The invention discloses a medical image classification method based on a capsule theory and PLSA routing. A PLSA probability model is introduced, a new capsule routing method is designed, and comparedwith consistent routing, this method uses a more reasonable soft clustering method to measure the consistency between capsules. Based on this, a medical image classification model is designed, whichcombines the capsule theory and PLSA routing, to better handle the task of medical image classification. The model framework consists of a convolutional neural network based on capsule and a fully connected network for image reconstruction. The convolutional neural network based on capsule uses PLSA routing to transfer the information between capsules, which not only can automatically learn the medical image features, but also can better find the size, position and direction of the features, and improve the classification accuracy. In order to improve the generalization ability of convolutional neural network based on capsule, full-connected image reconstruction network uses category capsule to recover original medical images.