CEU-Net: ensemble semantic segmentation of hyperspectral images using clustering

Abstract Most semantic segmentation approaches of big data hyperspectral images use and require preprocessing steps in the form patching to accurately classify diversified land cover remotely sensed images. These incorporate rich spatial neighborhood information exploit simplicity segmentability most common datasets. In contrast, landmasses world consist overlapping diffused classes, making weaker than what is seen To combat this issue generalize models more complex diverse datasets, work, we propose a novel flagship model: Clustering Ensemble U-Net. Our model uses ensemble method combine spectral extracted from convolutional neural network training on cluster landscape pixels. outperforms existing state-of-the-art methods gets competitive performance with without when compared baseline models. We highlight our model’s high across six popular datasets including Kennedy Space Center, Houston, Indian Pines, then compare them current top-performing