Rice Semantic Segmentation Using Unet-VGG16: A Case Study in Yunlin, Taiwan

Abstract

In this paper, Unet-VGG16 semantic segmentation network is proposed to segment rice regions in the aerial images of Yunlin, Taiwan. The experimental results show that different combinations of image bands and different image conditions affect the segmentation accuracy. With R-G-NIR bands as input and bright aerial images as the dataset, the Unet-VGG16 network yielded the best segmentation result, achieving a test accuracy of 0.91.