Insect detection in sticky trap images of tomato crops using machine learning

Abstract

As climate change, biodiversity loss, and biological invaders are all on the rise, the significance of conservation and pest management initiatives cannot be stressed. Insect traps are frequently used in projects to discover and monitor insect populations, assign management and conservation strategies, and assess the effectiveness of treatment. This paper assesses the application of YOLOv5 for detecting insects in yellow sticky traps using images collected from insect traps in Portuguese tomato plantations, acquired under open field conditions. Furthermore, a sliding window approach was used to minimize insect detection duplicates in a non-complex way. This article also contributes to event forecasting in agriculture fields, such as diseases and pests outbreak, by obtaining insect related metrics that can be further analyzed and combined with other data extracted from the crop fields, contributing to smart farming and precision agriculture. The proposed method achieved good results when compared to related works, reaching 94.4% for mAP_0.5, with a precision and recall of 88% and 91%, respectively, using YOLOv5x.