Research On Hyperspectral Image Classification Method Based On Deep Learning

With the development of society and the progress of science and technology,the development of hyperspectral remote sensing technology has brought new opportunities and challenges to the fields of precision agriculture,food safety,industrial sorting,cultural relics protection,criminal investigation and document inspection,camouflage identification,environmental monitoring,pharmaceutical medicine and so on.Compared with ordinary images,hyperspectral images have many data dimensions,large amount of data and too few labeled samples,which are not convenient for hyperspectral image classification.To solve these problems,supervised learning and unsupervised learning are commonly used.Deep learning is a new direction to solve this problem,and it is also a research hotspot in the field of machine learning in recent years.In this paper,hyperspectral remote sensing images are processed from two routes:spectral feature extraction and classification,spatial-spectral feature extraction and classification.The specific work is as follows:(1)In view of the fact that hyperspectral image data contains noise,band number and redundant data,and the classification effect of common classification methods is general,a hyperspectral image classification algorithm based on multi-layer feature extraction is proposed.Firstly,data preprocessing is carried out.Secondly,a new method of hyperspectral pixel mode input is proposed,which makes up for the appropriate zero of the pixel features of hyperspectral images and converts them into two-dimensional square matrices.Then,the generated two-dimensional square matrices are extracted and classified by PCANet.In order to obtain better classification accuracy,the influence of PCANet parameters on classification accuracy is discussed in detail.Experiments on Indian Pines and Pavia University datasets show that the proposed method has better classification accuracy and better performance than traditional methods.It is an effective and feasible hyperspectral classification method.(2)Traditional hyperspectral classification methods often only use the spectral characteristics of pixels,ignoring the characteristics of hyperspectral image "atlas in one"and not making full use of the spatial characteristics of hyperspectral images.In this paper,a deep learning hyperspectral image classification method based on space-spectrum combination is proposed.Firstly,the hyperspectral image data are preprocessed and dimension reduction operations are carried out.Secondly,the spatial feature information of hyperspectral image is extracted by LBP operator.Then the hyperspectral image is divided into blocks.Then the deep convolution neural network is used to learn the deep features.Finally,the multi-layer perceptron classifier is used to extract and classify the features.In order to verify the classification performance of the proposed algorithm,experiments on Indian Pines and Pavia University datasets are carried out and compared with traditional methods.The experimental results show that the proposed algorithm is superior to the spectral feature extraction algorithm only.It is an efficient and robust hyperspectral classification method.