Application of computer vision technology to the regurgitation behavior of fruit fly (Diptera: Tephritidae)

Abstract Fruit fly regurgitation contains a variety of behavioral information such as predation and defense. The study of regurgitation behavior in fruit fly helps to understand the intrinsic connection between other physiological behaviors of fruit fly,which is helpful for fruit fly-specific control and can significantly improve the quality and yield of fruits. In this paper, based on the existing network models, three different methods based on computer vision techniques are proposed to recognize fruit fly regurgitation, extract regurgitation spots and track the trajectong of fruit fly regurgitation. And the methods can be applied to other insect behavioral studies. The Top-1 Accuracy of I3D model in fruit fly regurgitation recognition registers 96.3 percent. The MIOU of the combination of Unet and CBAM attention mechanism in segmenting regurgitated spots can achieve 90.96 percent. Then we conducted threshold segmentation, using OpenCV to calculate the amount and area of regurgitation spots. The accuracy of Yolov5 in detecting fruit fly reached 99.8 percent. And combined with DeepSort model, it can track fruit fly accurately.