Detection of Lygus Bugs in Strawberry Using Deep Learning

Abstract

ABSTRACT Lygus bugs are significant insect pests of strawberries that cause substantial reduction in yield and quality of fruits. In California, tractor-mounted bug vacuums are used to mechanically control Lygus population in the field. Traditional methods to evaluate the performance of bug vacuum are tedious and time-consuming. Automated insect detection using machine learning is an effective approach for pest management which can overcome the drawbacks of conventional techniques and can be helpful to estimate the removed bugs by the vacuums. The objective of this research was to detect adult and nymph Lygus using deep learning in lab settings. A monochrome camera, an optical filter, and a light were used to capture videos of moving Lygus inside of a chamber. Two training models, MobileNet SSD v2 and Faster R-CNN ResNet-50 v1, were applied to recognize adult and nymph Lygus in images. The results showed that both models could accurately identify the Lygus classes (F1- score ≥ 0.80). This can be useful to assess the performance of bug vacuums and create a Lygus heatmap to monitor its population in the field.