High-Resolution Remote Sensing Image Classification through Deep Neural Network

Remote sensing in image processing is popular in urban monitoring, forest detection, and disaster Monitoring. The high-resolution satellite images are classified into their respective classes through their distinctive features. Innovation in image acquisition has played a critical role in the process of recognition. However, the geometric and photometric variations require the extraction of invariant features. This paper presents a robust strategy that can classify such high-resolution images, also in case of changes in geometry and photometry. The employed dataset consists of images located in the Headwater Region of China. The images of the database include variations in illumination, viewpoint, and scale. Robust and distinctive features collected from the fully connected layer of the DNN model are classified through a multi-class support vector machine. The Gaussian kernel type parameter of SVM is used for the classification in our experiments. The results show our proposed approach provides 93.8% classification accuracy, which is better than many recently reported works.