Tile-Based Semisupervised Classification of Large-Scale VHR Remote Sensing Images

Figure 3

Large VHR dataset used and its qualitative results. (a) Original image of size 6500 × 10400 with example training tiles (10 per class), (b) testing map (ground truth), (c) legend, (d) NN classification result (OA = 89%), (e) final map (OA = 92.2%), and (f) tiles corrected by semisupervised learning.

(a)

(b)

(c)

(d)

(e)

(f)