Intelligent Monitoring of Insects with Computer Vision Using Convolutional Neural Network-Based Deep Learning Techniques: A Systematic Mapping Study

Abstract

Highlights Systematic mapping study on insect monitoring with computer vision. Review of 114 relevant research studies. State-of-the-art of multi-object detection and tracking applied to insect monitoring. Challenges and possible solutions for insect monitoring. Abstract. Insect monitoring offers several benefits to agriculture and ecology, primarily as a tool for enhancing pollination services, managing pests, promoting the rational use of pesticides, and combating insect decline to restore abundance, diversity, and biomass. Traditional monitoring requires repetitive work that depends on specialists. Intelligent insect monitoring has the potential to overcome the shortcomings of conventional monitoring and make it more efficient. We consider that monitoring should detect and track insects of interest. This article provides a comprehensive overview of the current literature, highlighting gaps, trends, and research challenges that warrant further investigation. We conducted a systematic literature review following well-established methodological guidelines. We carefully selected 38 primary and 76 derivative studies as a sample of 2077 articles from diverse databases. We analyzed and categorized the articles to address seven research questions. The results show that 103 of the studies applied insect detectors and Multiple Object Tracking, 85 utilized insect datasets, 16 proposed hardware, and a single work addressed the energy cost of the hardware. Computer vision using deep learning techniques based on Convolutional Neural Networks presents a promising approach to overcoming the challenges of intelligent insect monitoring. Ultimately, this article can serve as a starting point for future studies, offering insights from a systematic literature review. Keywords: Bees, Computer vision, Deep learning, Insects, Intelligent monitoring, Multiple Object Tracking.