Image Classification for Soybean and Weeds Based on ViT
Jingxin Liang, Dong Wang, Xufeng Ling
Abstract
Abstract Abstracts. In this paper, ViT deep neural network based on self-attention mechanism is used in classification for images of soybean and weeds. Firstly, the overall image is split into multiple tiles; with each tile regarded as a word, the whole image is regarded as a sentence, which can be used for image semantic recognition by natural language processing technology. We designed a ViT network with sequence length of 50, embedded dimension of 384, and self-attention module layers of 12. With soybean weed classification dataset, the network is trained, verified and tested. Experimental results showed that ViT network is superior in classification on dataset of soybean and weeds, with excellent generalization capability.