Skip to main content
Fig. 6 | EURASIP Journal on Advances in Signal Processing

Fig. 6

From: Learning from label proportions for SAR image classification

Fig. 6

Classification accuracy using GL+LpcSVM for the SAR images of Tianjin (a) and Rosenheim (b) as a function of the iteration number of LpcSVM under different cell sizes. Cell size ranges 1 ×1, 100 ×100, …, 500 ×500. Note that cell size of 1 ×1 means the traditional SVM with pixel labeling (PL+SVM), making a horizontal reference line through different iterations. For Rosenheim, the maximum cell size is 300 ×300 as larger cell sizes will cause the land cover of water body (WB) disappear in the training set. Iterations of four will produce a stable value of classification accuracy

Back to article page