A Shifted Windows Detecting Method for Similar Plant Diseases Based on Transform
Conference: ISCTT 2022 - 7th International Conference on Information Science, Computer Technology and Transportation
05/27/2022 - 05/29/2022 at Xishuangbanna, China
Proceedings: ISCTT 2022
Pages: 4Language: englishTyp: PDF
Authors:
Wang, Fenmei (Institute of Intelligent Machines, Hefei Institutes of Physical Science, Chinese Academy of Science, Hefei, China & University of Science and Technology of China, Hefei, China & PLA Army Academy of Artillery and Air Defense, China)
Wang, Rujing (Institute of Intelligent Machines, Hefei Institutes of Physical Science, Chinese Academy of Science, Hefei, China & University of Science and Technology of China, Hefei, China)
Huang, Huanhuan; Wu, Suqin (PLA Army Academy of Artillery and Air Defense, China)
Abstract:
Plant disease is one of the main threats to crops, and the identification and detection of plant leaf diseases with high similarity is a key and complicated technology. The mainstream approach is to use the CNN architecture to deploy the model. Unfortunately, the disease effect is not ideal. Instead, this paper proposes a method to identify similar diseases using Shifted Windows detecting with transform architecture. In multi-head attention in the structure, we use cross-attention instead of self-attention to improve the detection of confusion brought by similar features. The experimental results show that the algorithm achieves an mAP of 35.7%.