Early control of spodptera litura lotus root pest using image processing

Pest and disease control is essential in agriculture fields, and efforts are made to control pests for each plant species. The lotus root, which is the subject of this research, is specifically grown in Tokushima. Tokushima’s production of lotus root is third largest in Japan. However, leaves and roots are affected by pests, disease and weather disasters including typhoons. Pest control is difficult in lotus root fields because the lotus root is cultivated in paddy fields and the leaves grow to two meters. A variety of pests and diseases have been identified as affecting lotus root crops, including aphids, <i>Thysanoptera</i>, <i>Spodoptera litura</i> and rot disease, <i>Corynespora cassiicola</i>. Early detection of Spodoptera litura is especially necessary because the larvae are difficult to detect and as the larvae grow older, pesticides become less effective. Therefore, this research focuses on the feeding scars of the larvae and enables early detection by detecting damage scars from UAV images taken in lotus root fields. The damage detection is carried out in a two-step process. First, we detect leaves by Mask R-CNN that is an instance segmentation model. Next, we extract damage marks area by U-Net that is a semantic segmentation model. Finally, usefulness is demonstrated by calculating accuracy in detecting damaged leaves detection.