Development of long-term night-vision video analyzing system for physical pest control

This paper presents a prototype of a long-term night-vision video analyzing system intended for labor-saving agriculture use. Recently, physical stimulation approaches using light, sonar, microwave, and vibration have been assessed for pest control. For accurate application of physical stimulus to pests directly, it is necessary to elucidate their ecology and habits. However, numerous habits remain unknown, especially those of nighttime. As a primary step for understanding the ecology of pests and analyzing their habits, it is necessary to identify metamorphoses such as extrication, pupation, and eclosion. For this study, we proposed a novel method combined with background subtraction and optical flows for the detection of eclosion, flying, and mating. Our measurement system consists of an incubator and two night-vision cameras. In the incubator, we released cabbage moths in the space surrounded by an acrylic plate. We prepared a dish containing sugar water as feed. Night-vision cameras installed to the upper part of the incubator recorded time-series images during daytime and nighttime for 15 days, switched artificially. The experimentally obtained results revealed that behavior patterns of eclosion and mating were stably detectable, except for overlapping of imagoes and cloudy conditions in the acrylic case. Moreover, results showed that eclosion occurred during nighttime. Motions changed remarkably according to environmental changes. Furthermore, cabbage moths moved only slightly before mating and slightly during mating. We obtained knowledge of basic tendencies of behavior patterns according to ground truth datasets to achieve labor saving for automatic pest control.