Classification Of Feature Recognition Based On Hyperspectral Image

Remote sensing images play an important role in military,national defense,urban planning,agriculture and so on,so the classification of images is also particularly important.high-resolution?high-space images provide people with more space and spectral information and various types of features,but those bring two Important Problems,include large amounts of data to hyperspectral images and mixed pixels caused by spatial resolution.the traditional method has been difficult to meet the needs of modern high-precision identification,while the traditional method to select a single feature to identify the features,rarely use multi-feature for feature recognition.Aiming at the demand of high-precision identification and the problem of double-high image recognition,Based on the existing foundation,this paper constructs three learning frameworks to improve the recognition accuracy of hyperspectral images.The research contents are as follows:First of all,propose to use stacked denoise autoencoder to extract deep feature of hyperspectral images,and link a classifier named dictionary pair learning.In view of the shortcomings of a single feature,extracte the deep features from spatial-spectral characteristics.The experimental results show that the proposed method has a good experimental effect.three sets of simulations are made.The experiment shows that the frame has high accuracy and high recognition effect.Secondly,A convolutional neural network model is designed to identify the image features,According to the characteristics of image data,convolutional neural model is trained.The characteristics of the model are not required to be manually extracted and simple to operate.The experimental results show that convolutional neural network has good effect on the recognition of hyperspectral imagesFinally,For mixed pixel problems,proposes subspace-based dictionary pair learning for Hyperspectral image.Subspace projection method to better characterize noise and highly mixed pixels.The dictionary pair learning is an improvement to dictionary learning,which uses class tag information to enhance the recognition of coding.At the same time,the integration of spectral and spatial features for classification studies.Based on different algorithms,five groups of simulation experiments to do comparative analysis.Experimental results show that the proposed method better than the currently popular algorithm in term of time and accuracy.