33 results about "Multispectral image classification" patented technology

The invention discloses a high spectral image classification method based on NSCT transformation and DCNN. The objective of the invention is to solve the problem that texture details and directivity information of to-be-classified high spectral images cannot be sufficiently excavated in the prior art. The method comprises steps of inputting a high spectral image; carrying out NSCT transformation; carrying out normalization and block taking operation on the transformed stereo blocks; randomly selecting training, verification and test sample sets in a sample set; constructing a depth convolution neural network and setting network super-parameters; training the network; inputting the test samples into the network to obtain actual classification tags and drawing terrain classification result graph; and comparing the classification tags and the reference tags of the test samples, calculating classification evaluation indexes, drawing loss curve graphs of the training and verification samples of increasing along with the iteration times, thereby finishing the terrain classification. According to the invention, more texture details, directivity information and space information of the high spectral image are kept; the classification is quietly precise; and the method can be applied to meteorology and environmental monitoring, land utilization, urban planning and disaster prevention and reduction.

The invention discloses a hyperspectral image classification based on gradient lifting decision tree and semi-supervise algorithm integration in order to solve the technical problem that hyperspectral image classification based on active learning and semi-supervise learning is low in classification precision. Hyperspectral image classification includes the steps that firstly, hyperspectral image data is input; secondly, features of sample points are extracted; thirdly, parameters of a gradient lifting decision tree classifier are trained; fourthly, massed learning sample points are classified; fifthly, the confidence degree of the sample points are assessed; sixthly, the sample points are screened through sparse representation; seventhly, a marked training set is updated; eighthly, a classification result is output. Assessment is conducted on the confidence degree of the unmarked sample points through the prediction result of the classifier and sparse representation, according to the confidence degree of the unmarked sample points, the sample points are divided into two sets for different kinds of processing, burdens for manual marking are reduced while classification precision is improved, and hyperspectral image classification can be used in the fields of geological survey, atmospheric pollution and like.

The invention discloses a spatial spectrum attention hyperspectral image classification method based on Octave convolution, and solves the problems of large class spacing, small different class spacing and low classification accuracy in the prior art. The scheme is as follows: inputting images to be classified and preprocessing data, dividing a training set and a test set, constructing an Octave convolutional neural network, determining a loss function of the Octave convolutional neural network, training and updating the Octave convolutional neural network, testing the data of the test set, and completing hyperspectral image classification. According to the method, Octave convolution operation is used to reinforce feature representation, and a spatial attention mechanism and a spectral attention mechanism are introduced, so that the network can more accurately find an area which is more beneficial to classification and contains more comprehensive and detailed information. The method ishigh in classification precision and strong in robustness, and can be applied to analysis and management of hyperspectral image data.

The invention provides a high-spectral image classification method based on an improved learning model, and relates to the technical field of high-spectral images. The method comprises that one or more convolution layer and pooling layer form an extracted image characteristic alternatively, the obtained image characteristic is extracted for multiple times randomly to form different classificationmodels, the different classification models are used to carry out classification, and classification results are voted to obtain a final classification result in the principle that the minority submitto the majority, and the deep learning network model is constructed. The high-spectral image data to be classified is reconstructed, space spectrum information of high spectrum images is maintained,the sparsity of the data is increased, and thus, the high spectral image is easy to process by the classification model. The method includes fewer calculation parameters, is higher in classification accuracy, and can be used to classify natural objects in the high spectral image finely.

The invention belongs to the field of hyper-spectral intelligent perception, and particularly discloses a spatial spectrum fusion hyper-spectral image classification method based on a three-dimensional deep residual network, which comprises the following steps: S1, generating candidate frames by using a sliding window method, and generating a plurality of windows; S2, randomly dividing the windowinto a training set and test set data; S3, training a three-dimensional depth residual network (3D-CNN) based on the hyperspectral data in the training set; S4, inputting a test set sample into the classification model, The hyperspectral image classification method has the advantages that the characteristics of the input data are extracted and predicted, the spectral characteristics and the spatial spectrum characteristics of the hyperspectral image are extracted at the same time, the classification precision of the hyperspectral image is further improved, a residual network structure is introduced, and the problem of learning degradation in a traditional hyperspectral classification neural network is solved. The hyperspectral image target classification method is clear in structure and easy to implement, the structural characteristics of the hyperspectral image can be fully utilized, and the hyperspectral image target classification precision is remarkably improved while the calculation time is shortened.

The invention discloses a multi-spectral image classification method based on threshold self-adaption and a convolutional neural network. The method comprises the following steps of inputting multispectral images of different time phases and different wave bands of satellites to be classified, and carrying out normalization on all pixels of a marked portion of a same wave band image in all cities; stacking selected 9 wave bands into one image and taking the image as a training data set; constructing a classification model based on the convolutional neural network, using the training data set to train the classification model so as to acquire a probability model based on OSM, using the model and a confidence coefficient strategy to adjust a softmax output result and acquiring a final classification model, and finally, uploading a test result to an IEEE website so as to acquire classification accuracy. By using the multi-spectral image classification method, characteristics that there are a lot of wave bands, a data volume is large and there are a lot of information redundancy in the multispectral images are fully used so that a problem that surface features of complex types are difficult to classify is solved, classification accuracy is increased, an error dividing rate is reduced and a classification speed can be increased.

The invention discloses a hyperspectral image classification method based on spatial information enhancement and a deep belief network and mainly solves a problem that the accuracy improvement of the hyperspectral image classification in the existing technology is not obvious. The technical scheme is as follows: the method comprises the following steps of 1) carrying out normalization and band selection on an original hyperspectral image to obtain a hyperspectral image with a reflectance value of 0 to 1; 2) carrying out spatial information enhancement on the hyperspectral image through band grouping and ground and guided filtering; 3) constructing a deep belief network model according to the features of the hyperspectral image after spatial information enhancement; and 4) carrying out model training on the hyperspectral image after spatial information enhancement and utilizing the obtained model to carry out category prediction in order to obtain a classification result. According to the method, the spatial information of the hyperspectral image is effectively enhanced without losing the original spectral information, the classification accuracy is remarkably improved, and the method can be used for environment, climate, agriculture and mineral exploration.

The invention relates to a 3DCNN (three-dimensional convolutional neural network)-based high-spectral image space spectrum combined classification method. According to the feature that high-spectral image data is of a three-dimensional structure, the three-dimensional convolutional network suitable for a high-spectral image is constructed and high-spectral image space spectrum combined classification is completed. First of all, data blocks within a certain neighborhood scope, taking a pixel to be classified as a center, are extracted from an original high-spectral image as initial space spectrum features, and through combination with a label of the pixel to be classified, the constructed 3DCNN is trained. Then, by use of the trained 3DCNN, the high-spectral image space spectrum combined classification is completed. The method has the following advantages: 1, the problem of need of complex processing of spectrum space dimension reduction or compression in the prior classification arts is solved; 2, the 3DCNN suitable for the high-spectral image data of the three-dimensional structure is constructed, rich information of the high-spectral image is fully utilized, and the trouble of manual set of features in advance is omitted; 3, the 3DCNN-based high-spectral image space spectrum combined classification method enlarges the application scope of depth learning and also provides a new approach for high-spectral image classification; and 4, the high-spectral image classification precision is improved.

The invention discloses a hyperspectral image classification method based on spatial feature iteration. The hyperspectral image classification method comprises the steps of: calculating spectral signatures di of ground feature categories according to hyperspectral sample data; setting a category target set, a background end member set and a color constraint matrix; defining a classifier Tk capable of classifying multiple categories simultaneously by utilizing background end members U of a hyperspectral image, the constraint matrix C and an image autocorrelation inverse matrix R<-1>, extracting initial classification results of all category targets, extracting a spatial feature {Tk(iG)}of the classification result of each category target, and then feeding back and superposing the spatial features {Tk(iG)} into the hyperspectral image to be classified for carrying out fusion of the spatial features and inter-spectrum features; classifying multiple categories simultaneously by adopting an iteration mode until a set iteration stopping condition is reached; and marking classification results by using different colors. The hyperspectral image classification method effectively utilizes the method of fusing the spectral statistical features and the iteration spatial features to perform hyperspectral image category feature judgment, and gradually improves the accuracy of hyperspectral image classification.

The present invention discloses a hyperspectral image classification method based on a deep integrated residual network. Defects are overcome that human selection of weak classifiers and design of anintegration method are complex and time-consuming, the computation process is tedious and a classification result has phenomena of the same object with different spectra characteristics and differentobjects with the same spectra characteristic caused by a semi-supervised training mode in the prior art. The implementation of the method comprises the steps of: (1) inputting multispectral images; (2) performing ground object target removal normalization processing of images of each wave band of each multispectral image; (3) obtaining a multispectral image matrix; (4) obtaining a data set; (5) establishing a deep integrated residual network; (6) training the deep integrated residual network; and (7) performing classification of the test data set. The multispectral image classification methodbased on the deep integrated residual network has complete multispectral image features for learning, is more concise and much clearer in process, allows a classification effect to be more accurate, and can be used for classification of hyperspectral images.

The invention provides a high-spectral image classification method based on discrimination and robust multi-feature extraction. Different spatial spectral features in high-spectral images are projected into a common low-rank discrimination subspace, and significance of the different features is remained. The method comprises that the features are assigned with different weights according to the significance; a low-dimensional subspace with robust and discrimination features is learned; the different features are projected to the low-dimension subspace; and a support vector machine is used to classify the new features. The dimensions of the features are reduced, noise robustness is realized, noise interference in the high-spectral image can be defended, the significance of different features can be maintained, the obtained features has discrimination characteristic via exiting labels, and the classification effect is improved effectively.

The invention discloses a hyperspectral image classification-oriented data adaptive activation function learning method, which is used for solving the technical problem of poor accuracy of an existinghyperspectral image classification method. According to the technical scheme, spectral information and spatial information of hyperspectral data are utilized, and hyperspectral image classification is carried out by learning an activation function based on specific data. By analyzing a common activation function, it is found that the activation function can be represented in the mode that activation mapping is multiplied by input features. Therefore, activation mapping is learned by constructing a two-layer neural network. The neural network learns from two aspects of channels and spectrums,and activation functions conforming to the characteristics of the channels and the spectrums are constructed respectively, so that the image classification accuracy is improved. Tests show that on anIndian Pines data set, the classification accuracy of the activation function based on the channel and the spectrum is averagely improved by 2.17% and 4.03% respectively.

The invention discloses a discriminant Gabor filtering method for hyperspectral image classification, which comprises the following steps: reading hyperspectral image data and preprocessing; decomposing the standard three-dimensional Gabor filter into eight sub-filters; taking the CCS component (spatial low-pass, spectral dimension band-pass) of one of that sub-filters to construct a discriminantGabor filter; using different frequency amplitudes, different frequencies empty-spectral space direction, and discriminant Gabor filters at different scales to filter hyperspectral images respectively; taking the amplitude as the features, and obtaining the feature blocks under different parameters; connecting and merging feature blocks with different frequency parameters (frequency amplitude anddirection) at the same scale; selecting the scale parameters by a cross-validation method, and obtaining the optimal scale parameters, and using the features under the optimal scale parameters as thefinal discriminant Gabor features; finally, using a support vector machine classifier or least square classifier to classify the discriminant Gabor features of hyperspectral images. The method effectively improves the classification accuracy of the hyperspectral image.

The invention provides a fused panchromatic image and multispectral image classifying method and device. The fused panchromatic image and multispectral image classifying method comprises the steps of acquiring panchromatic images and multispectral images in a region to be detected, segmenting the panchromatic images to obtain over-segmentation bodies of the panchromatic images, merging the over-segmentation bodies of the panchromatic images to determine segmentation body units, classifying the multispectral images according to the segmentation body units to obtain classifying labels of the multispectral images, wherein each segmentation body unit comprises at least one over-segmentation body. By means of the fused panchromatic image and multispectral image classifying method and device, space information of the panchromatic images can be utilized in the multispectral image classifying process, the space information of the panchromatic images and the space information of the multispectral images can be fully and effectively utilized, the dependency on spectral information of image classification can be reduced, classifying accuracy can be improved, and the classifying labels can be clear.