An Efficient Application of Machine Learning for Object-Based Classification Using Sentinel-1 Data: For Complex Environment

Object-Based Image Analysis (OBIA), synonymous to Geographic Object-Based Image Analysis (GEOBIA), assesses the characteristics across spatial, spectral and temporal scales of the image-objects which are separated from the images. In contrast to pixel-based image analysis, which assigns category to each pixel of the image, OBIA considers the neighboring pixels and segments it to form “Meaningful objects” (Super-pixels). OBIA in satellite images like Sentinel-1, Sentinel-2, SAR have various applications in remote sensing like forest exploration, Land-cover monitoring, agriculture, Disaster management, etc. Although OBIA poses some challenges while dealing with complex environment, such as mixed classes, irregular shapes,etc. thus, there is a requirement of a robust approach which can effectively distinguish between these objects and classify it. This paper aims to apply Mean-Shift clustering to extract the objects and then classify the land cover types using Random forest and Support Vector Machine (SVM). Finally results are compared and it is concluded that the integration of Machine learning approaches with the use of OBIA can give impressive results in classifying various land covers with complex environment in which Random Forests shows better results than SVM.