Supervised remote sensing image segmentation using boosted convolutional neural networks

In this paper, a region segmentation technique for remote sensing images using a boosted committee of Convolutional Neural Networks (CNNs) coupled with inter-band and intra-band fusion, is proposed. The vast heterogeneity in remote sensing images restricts the application of existing segmentation methods that often rely on a set of predefined feature detectors along with tunable parameters. Therefore, it is highly challenging to design a segmentation technique which could achieve high accuracy while simultaneously maintaining strong generalization particularly for visual data with improved spatial, spectral, and temporal resolutions. The proposed method is a fusion framework consisting of a set of thirty boosted networks that derive individual probability maps on the location of region boundaries from the different multi-spectral bands and combines them into one using an averaging inter-band fusion scheme. The boundaries are then thinned, connected, and region segmented using a morphological intra-band fusion scheme. Qualitative and quantitative results, on publicly-available datasets, confirm the superiority of the proposed segmentation method over existing state-of-art techniques. In addition, the paper also demonstrates the effect of some variations in design-choices of the proposed method.