Smart insect trap: insect identification and classification using iNaturalist and ChatGPT

Abstract

Insect identification is crucial for sustainable agriculture and pest control, as it helps determine the pest type, reduces human labour, and decreases reliance on pesticides. However, there are a number of challenges associated with AI-based insect identification and classification, such as the lack of reliable data sources, imbalanced datasets and difficulty of annotation. Our paper offers a practical solution to these challenges using online platforms. Specifically, we coordinate multiple AI tools in an agentic-AI-style workflow: iNaturalist's computer vision model identifies the insect, ChatGPT determines whether it is a pest and the result drives a low-cost robotic trap. This approach is particularly accessible for non-experts, as it does not require extensive knowledge of machine learning or computer vision, and each component can be upgraded as better AI tools emerge. The system was further validated through field testing using images collected under real agricultural conditions. The results show our method is able to identify and differentiate insect species with sufficient accuracy. Our approach would be more accessible, reliable and efficient than traditional methods and can be used to identify a wider range of insect species.