A Deep Learning Approach Combining Super-Resolution and Segmentation to Identify Weed and Tobacco in UAV Imagery
Fan Zhao, Jirui Huang, Yongying Liu, Jiaqi Wang, Yijia Chen, Xinlei Shao, Bangzhang Ma, Dianhan Xi, Mowen Zhang, Zhengyue Tu, Mengya Wu, Qingyang Wu, Yulun Chen, Yinyin He
Abstract
Images of tobacco fields collected by drones suffer from Low-Resolution (LR), while the visual similarity between tobacco and weeds exacerbates the difficulty of tobacco segmentation. This study proposes a method for tobacco segmentation based on Super-Resolution Reconstruction (SRR) and semantic segmentation. We designed different strategies of super-resolution (SR) algorithms to enhance tobacco field images, followed by inputting the enhanced images into a semantic segmentation network to achieve precise segmentation of tobacco, weeds, and soil. The application of SRR results in a significant improvement in segmentation accuracy over LR images. Experimental results demonstrate the significant effectiveness of combining drone and SR technologies in tobacco and weed segmentation.