Description:
Abstract The stored-grain pests caused serious economic losses in the process of grain storage. So it is important for us to know the number and species of stored grain pests accurately as soon as possible. Then, take appropriate measures to reduce the economic losses. But current research of grain pest detection has two problems. The background of data set doesn’t contain any grain, so the results couldn’t apply to detect pests in the actual situations. The other problem is the methods based on pest trap can’t actually reflect situations of the surface of the whole granary. This paper proposes a system of pest detection and counting to solve the problems. This system main consists of two parts that are deep learning target detection model called YOLOV4 and a small car with camera. The model has been trained is embedded in raspberry pi to run. Then, the raspberry pi and a camera are carried by the small car is controlled by Micro Control Unit named STM32F407. In order to simulate the environment of the granary, take 2 typical stored-grain pests to wheat, Red Flour Beetle and Rice Weevil as the research object. The camera on the small car can take photos of the pests for the model to make data set for training. By training and testing the stored-grain pest images, the results demonstrate that the proposed system could reflect the real pest situation in the granary