Integrated Deep Learning Framework for Insect Detection and Classification in Video Streams

In the context of contemporary environmental concerns, insects have emerged as significant challenges due to their role in disease transmission and their adverse impact on agricultural productivity. Addressing these multifaceted issues demands innovative solutions. The integration of computer vision and deep learning emerges as a potent approach, offering pathways to resolve challenges within the realm of insects. This paper introduces an innovative framework for the detection and classification of insects utilizing computer vision and deep learning. In the initial phase, a computer vision algorithm is employed for real-time insect detection within video streams. Subsequently, a classification algorithm is applied to ascertain the specific insect type detected in the preceding stage. The empirical findings, derived from experiments conducted on the synthetic insect Village dataset, highlight the superiority of the YOLOv8 detection algorithm over both Faster R-CNN and YOLOv4. Notably, YOLOv8 achieves remarkable mAP50 and mAP50-95 scores of 0.962% and 0.804%, respectively. Additionally, the species classification task demonstrates the superiority of VGG16 over EfficientNet-B3, boasting an impressive accuracy rate of 0.993%.