Performance Analysis of Vegetation Area Classifications in Satellite Images Using Machine and Deep Learning Approaches

At present, the vegetation area around the world is shrinking due to the development of construction area in both urban and rural areas. It is very important to expand the present vegetation area to meet the food requirements of all people in world. In order to cope with this aspect, the present vegetation areas should be detected. In this paper, the vegetation areas in remote satellite images are detected and segmented using machine learning and deep learning algorithms. The machine learning algorithm Support Vector Machine (SVM) consists of preprocessing, feature extraction and classification modules where the deep learning algorithm consists of data augmentation and Convolutional Neural Networks (CNN) classification module. In this paper, the conventional CNN architecture is modified in this paper as the novelty in order to improve the classification accuracy of the proposed satellite image system. The segmented vegetation area is compared with manually segmented images in order to evaluate the performance of the proposed system. The developed CNN architecture produces features itself in each Convolutional layers. The CNN based vegetation area segmentation method achieves 96.03% of SEN, 98.12% of SPE and 98.07% of ACC and SVM based vegetation area segmentation method achieves 94.12% of SEN, 96.67% of SPE and 97.01% of ACC.