Construction and evaluation of an AI system for tracking cricket behavior under conditions of bright and dark lighting

Abstract

Since 2013, the Food and Agriculture Organization of the United Nations published a publication that embraced the opportunity to use insect for food and feed, to raise the profile of insects and further promoting diversity of diet, sustainable practices, and food security, insect production has attracted attention as a potential food source, and advanced methods have been developed. Among insects, crickets are of particular interest because they are omnivorous and easy to mass produce. Ultimately, we aim to build a system in Japan that can mass-produce crickets using food waste. For this purpose, a cricket feeding system that utilises food loss is needed to support the development and commercialisation of new insect foods along with renewable energy and smart production methods to reduce labour requirements. To improve productivity and develop smart production, we should understand the ecology of crickets (focusing on their behaviour in terms of diet and water consumption) with the premise of mass production. However, relatively few studies have considered these factors in detail. We consider that this information could enable considerable improvement in the productivity of cricket production systems. Therefore, in this study, we constructed an Artificial Intelligent system designed to capture ecological information on crickets using a You Only Look Once version 5 model for mass production. The proposed system is relatively inexpensive, has high recognition accuracy and real-time performance, can be configured for different lighting conditions, and can recognise crickets even under dark conditions. The results show that this system can be implemented on prototype smart cricket farms and can be integrated with mass production systems. It is robust to bright and dark conditions with high accuracy and real-time framerates, with mAP values of 0.903 and 0.921 and framerates of 20.20 and 20.96 frames, respectively.