Hyperspectral Remote Sensing Image Processing System Based On Full Convolutional Network

In recent years,the deep learning has been widely used in image processing.The Fully convolutional networks(FCN)improve the architecture of Convolutional Neural Network It produces a dense prediction,of each pixel to make a pixelwise prediction for image classification.FCN can address the problem of semantic image segmentation efficiently.This paper researched on hyperspectral remote sensing image using the advantage of FCN to process the hyperspectral remote sensing image in order to extract the feature of the image and making use of the feature to deal with image classification.The works of this paper are as follows:(1)The method of hyperspectral remote sensing image feature extraction based on the fully convolutional networks is implemented.Owing to the traditional feature extraction algorithm of hyperspectral remote sensing image can-not combine the spectral information and spatial information.In order to solve this problem,this paper proposes a method of hyperspectral remote sensing image feature extraction based on the fully convolutional networks.The implemented method combines the spectral feature and spatial feature for image classification.The experimental results show that the implemented method can yield state-of-the-art classification results for different hyperspectral datasets.(2)The method of hyperspectral remote sensing image feature extraction based on the combination of convolutional layers is implemented.The influence of different combination of convolutional layers and the features on the classification results are deeply explored.Based on the feature extraction algorithm of hyperspectral remote sensing images using full convolution network,some contrast experiments are conducted for the purpose of researching the influence of different combinations of convolutional layers and the features on the classification results.The experimental results show that the features obtained from different combination of convolution layers have little effect on the classification result of hyperspectral remote sensing images,and the combination of features can improve the classification effect of hyperspectral remote sensing images.(3)Considering the feature extraction algorithm of hyperspectral remote sensing images based on full convolution network and the following sparse representation classification,this paper designs a software system for hyperspectral remote sensing images classification.This software system contains the overall process of hyperspectral remote sensing images classification.We can directly and clearly contrast the results of images classification on different datasets with different parameter setting and features.