[D] Animal husbandry, sanitation, house pests

[D-33] 粘着トラップに捕殺されたサシバエの機械学習を用いたモニタリング

◯Satoshi Kamitani¹ (1. Kyushu University)

The stable fly, Stomoxys calcitrans, is a serious pest that affects the health of cattle by the transmission of BLV. Monitoring the density of flies is crucial for evaluating the necessity and effectiveness of IPM. Sticky traps are commonly used for monitoring, but manual identification is required. However, with advancements in machine learning technology, it has become possible to perform insect identification. This study aims to distinguish and automatically count the number of stable flies captured on sticky traps, differentiating them from other insect species. Machine learning techniques, specifically YOLOv5, YOLOv8, and SAHI, were employed. Using 629 training images and 96 validation images of single images, the identification results for stable flies were mAP@0.5=0.99. This indicates high precision and recall. In the result with YOLOv8 and SAHI, without downsizing images, it was high precision and recall. This suggests that it is possible to identify and automatically count the number of Stomoxys calcitrans using these techniques.