How Useful is Region-based Classification of Remote Sensing Images in a Deep Learning Framework?
This work addresses classification accuracy in remote sensing images, but it is incremental as it compares existing segmentation methods without introducing new techniques.
The paper investigates using segmentation algorithms as a preprocessing step for classifying remote sensing images in a deep learning framework, finding that superpixel algorithms improve classification accuracy by providing homogeneous and compact segmentation.
In this paper, we investigate the impact of segmentation algorithms as a preprocessing step for classification of remote sensing images in a deep learning framework. Especially, we address the issue of segmenting the image into regions to be classified using pre-trained deep neural networks as feature extractors for an SVM-based classifier. An efficient segmentation as a preprocessing step helps learning by adding a spatially-coherent structure to the data. Therefore, we compare algorithms producing superpixels with more traditional remote sensing segmentation algorithms and measure the variation in terms of classification accuracy. We establish that superpixel algorithms allow for a better classification accuracy as a homogenous and compact segmentation favors better generalization of the training samples.