A Method of Establishing a Synthetic Dataset for Stored-Grain Insects

This paper presents a method to establish the image dataset for stored-grain insects based on image composition, for the development of stored-grain insect recognition. The proposed method comprises 3 parts: 3D modeling of insects, the dynamic generation equation for synthetic image of stored-grain insects, and a material conversion neural network. As a result, a synthetic image dataset, named VirtualInsect, is established for adults of insects from 13 species 11 families in main stored grains (paddy rice, wheat and corn) in China, which is the most diverse in species of insects at present.