Abstract
The investigation of the complicated underground life via automatic technique is in high demand in recent days. Using Convolutional Neural Network (CNN) to detect soil in- vertebrates is an interesting approach, although most studies on the topic have focused on other solutions. The creation of state-of-the-art technique through this work will be a significant step in soil ecology, bio-science and agriculture in effectively exploring the different types of invertebrates, their behaviors and interactions. In this paper, generating and annotating images containing seven classes of invertebrates is firstly presented. Then various automatic detections of the invertebrates using YOLOv5 algorithm on these images are performed and evaluated. Index Terms—Soil fauna, invertebrate detection, small object detection, invertebrate dataset, YOLOv5, cluttered background