Research On Hyperspectral Image Recognition Of Feature Based On Machine Learning

With the rapid development of remote sensing data acquisition technology and the increasing demand of various industries,remote sensing data has changed from wide band to narrow band,and hyperspectral technology has entered a period of rapid development.The spectral information of hyperspectral has the characteristics of narrow range,multiple and continuous bands,so that the features of ground objects appearing in the narrow band of absorption/reflection peak can be significantly different.Pattern recognition,machine learning and deep learning have also been gradually introduced into remote sensing detection technology at different stages of development for remote sensing data processing and analysis.However,traditional hyperspectral image ground object classification often make insufficient use of spatial information,and spatial information mining based on experience has great randomness.Because the machine algorithm has the advantage of strong automatic feature learning.This paper compares the advantages and disadvantages of different machine learning algorithms in the classification of hyperspectral image features,so as to provide scientific basis and practical reference for the recognition and classification of features.:(1)The principle,process and development of hyperspectral remote sensing characteristics and hyperspectral image classification are reviewed in detail.(2)The gated neural network in deep learning is applied in hyperspectral image classification to extract spatial information in the mode of sequence information.Firstly,the hyperspectral image is denoised and the band is selected,then the gated neural network is used for classification,and spatial information is extracted by sequential information pattern.Results show that the gated neural network has good performance in the classification of hyperspectral images.The classification accuracy is better than support vector machine,random forest,long and short time memory network and convolutional neural network.(3)A fusion model of three-dimensional convolutional neural network and gated neural network was proposed.Firstly,three-dimensional convolution is used to extract the features of hyperspectral images,and the problem of high correlation between adjacent bands of hyperspectral images is solved by dilated convolution and stride-length sliding,so as to obtain the spatial and spectrum fusion features.Then the gated neural network is used for classification and recognition to avoid the destruction of spatial information by the full connection layer.The results show that the classification accuracy of multimodel fusion is better than that of single convolutional neural network mode.