Behavioral Analysis of Insects Pollination Effects: A Precision Agriculture Approach

Abstract

In the context of urgent environmental concerns, insects have gained significant attention d ue to their impact on agricultural productivity, especially in relation to pollination processes. The study of insect behavior and its implications for pollination presents numerous challenges including the detect of the small insects, classify their species and study their behavior and their effects on pollination using manual methods cause an inefficient result that require multifaceted a nd innovative solutions. Utilizing computer vision and deep learning techniques offers a robust approach to address these challenges. This paper presents a framework that employs computer vision for the detection, classification, a nd tracking of insects within video streams. Specifically, the framework utilizes several computer vision approaches, including Yolo8, RTDETR, DETR with ResNet-50 backbone, and DETR with a ResNet-101 backbone. Empirical evaluations of these models are conducted using the Spatial Monitoring and Insect Behavioral Analysis Dataset. The results demonstrate that RTDETR outperforms other models, achieving remarkable mean Average Precision (mAP-50) scores of 84.1%. the YOLOv8 model secured a commendable 74.5% mean Average Precision (mAP-50), which earned it the second rank among the tested models.