Hyperspectral image classification based on adaptive segmentation

摘要: Abstract Object-based hyperspectral image classification (OBHIC) converts the basic unit from ‘pixel’ to ‘object’ by segmentation, in order take advantage of spatial distribution law geographical substances, as well increase performances. However, it involves problem scale selection, i.e. segmentation parameters are set manually empirical values. In this paper, a novel OBHIC algorithm based on adaptive is proposed. Here, images (HSIs) first segmented through new scheme with constraint ability, and thresholds for calculated adaptively utilizing training samples. And then K-nearest neighbor (KNN) applied classify centers each region after segmentation. addition, semisupervised idea, semi-known samples obtained further improve performance. Experimental results presented two HSI datasets. For different HSIs, consistent ones, developed has achieved good results, thus demonstrating strong robustness algorithm. Indian Pines AVIRIS sensor, Overall Accuracy (OA) kappa 95.13% 0.9444 respectively 10% samples, Pavia University ROSIS OA 95.52% 0.9416 2% performance still maintained small number