Remote Sensing Data Classification Using A Hybrid Pre-Trained VGG16 CNN- SVM Classifier
Nyan Linn Tun, Alexander I. Gavrilov, Naing Min Tun, Do Minh Trieu, Htet Aung
Abstract
In recent years, deep learning techniques have been improved to classify geographical information by assigning remote sensing images pixels. CNN models can fix the feature learning techniques in the field of visualization systems. This paper proposed a hybrid pre-trained VGG16 -convolutional neural networks (CNNs) - SVM classifier models. VGG16 conducts the features extraction from the input remote sensing data, and SVM classifier solves the classification output based on the CNN output feature maps. Our proposed model can play its neural network layers with a novel feature extraction strategy to achieve good classification accuracy over high-resolution remote sensing data. Classification experience is performed on the two remote sensing public datasets (UC Merced Land and RSSCN7), using high computational performance support that achieved reliable classification results within the shortest time.