Novel nocturnal insect pest monitoring for sustainable crop protection using ensemble augmented deep learning classification

Vine weevil, Otiorhynchus sulcatus F. (Coleoptera: Curculionidae), is an economically important pest of soft fruit and ornamental crops globally. Its management has historically relied on broad-spectrum synthetic insecticides, but this has shifted toward integrated pest management compatible methods such as entomopathogenic nematodes and fungi that target soil-dwelling larvae. These methods require reliable pest monitoring tools to be practically effective and economically viable. Existing monitoring methods rely on detecting the nocturnal adult weevils as a proxy for larval presence, however, these are unreliable and time-consuming to implement. This may be addressed by developing an identification algorithm for adult weevils. Here we present results that show improved machine learning models can identify adult vine weevils under laboratory and semi-field conditions. Specifically, we employ a lightweight network model and use ensemble enhancement techniques to address potential issues such as color variations, occlusions, and deformations in the data labels. The proposed framework strategically integrates a lightweight network model with adaptive ensemble augmentation mechanisms to comprehensively address three core data challenges: (1) chromatic variance under varying illumination conditions, (2) partial occlusion from pest aggregation, and (3) morphological deformation during specimen collection. This is the first report of such technologies specifically developed for a nocturnal insect pest. It demonstrates the feasibility of an automated monitoring approach, which could benefit growers as it will provide more timely information about pest populations in their crops and better inform management decisions.