A lightweight and enhanced model for detecting the Neotropical brown stink bug, Euschistus heros (Hemiptera: Pentatomidae) based on YOLOv8 for soybean fields

Insect pest detection and monitoring are vital in an agricultural crop to help prevent losses and be more precise and sustainable regarding the consequent actions to be taken. Deep learning (DL) approaches have attracted attention, showing triumphant performance in many image-based applications. In the adult stage, this research considers detecting a vital insect pest in soybean crops, the Neotropical brown stink bug (Euschistus heros), from field images acquired by drones and cellphones. We develop and test an improved YOLO-model convolutional neural network (CNN) with fewer parameters than other state-of-the-art models and demonstrate its superior generalization and average precision on public image datasets and the new field data provided here. Considering the proposal's precision and time of response, the possibility of deploying this technology for automatic monitoring and pest management in the near future is promising. We provide open code and data for all the experiments performed. •An improved YOLOv8 model tailored for detecting insects in field images.•Ablation experiments with new modules showing the effectiveness of their inclusion.•Tests with a public dataset and with novel field-collected images of soybean fields.•A dataset of NBSB images in soybean fields for testing.