AUTHOR=Yang Shuai , Xing Ziyao , Wang Hengbin , Gao Xiang , Dong Xinrui , Yao Yu , Zhang Runda , Zhang Xiaodong , Li Shaoming , Zhao Yuanyuan , Liu Zhe 

TITLE=Classification and localization of maize leaf spot disease based on weakly supervised learning

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 14 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2023.1128399

DOI=10.3389/fpls.2023.1128399

ISSN=1664-462X

ABSTRACT=Accurate identification of disease types and location of susceptible areas are important aspects of intelligent monitoring of crop production, as well as the basis for making plant protection prescriptions and automatic and precise application of drugs. This paper establishes a set of six types of field maize leaf image data collections based on images collected from Shangzhuang in Beijing, Hebi city in Henan Province, Dezhou city in Shandong Province, and open source datasets, and proposes a framework for maize leaf disease classification and location that combines lightweight convolution neural network MobileNetV2 with LayerCAM, an interpretable AI algorithm. The method achieves 98.62% recognition accuracy on the field maize leaf image data set. Moreover, with only image-level annotations, the mIoU of the lesion coverage and the actual coverage of the lesion reached 53.26%, which proves that the Interpretable AI algorithm is feasible for maize leaf spot recognition. This weakly supervised learning method, which combines the deep learning model with the visualization technology, increases the explainability of the deep learning model, and achieves the successful positioning of maize leaf infected areas by only weak supervised learning. The classifier can be deployed on mobile phones, smart farm machines, and other devices for intelligent monitoring and plant protection of crop diseases, and provides a reference for in-depth learning for crop disease research.