Towards Leaf Disease Recognition from Individual Lesions Using Deep Learning Techniques

Leaf disease recognition using image processing techniques is presently an active area of research. In recent years, most studies have focused on the use of deep learning techniques for crop disease recognition as these models have consistently outperformed shallow classifiers. When used to classify crop diseases from images taken under controlled lab conditions, deep learning models have achieved near perfect recognition accuracies. However, when used with images captured under field conditions, the deep learning models’ performance dropped considerably. Research showed that complex illumination and background conditions are mainly responsible for this decline in performance. Subsequent studies demonstrated that classifying images of individual lesions rather than images of whole leaves improved disease recognition accuracy while at the same time allowing for the detection of multiple infections presenting on the same leaf. Latest studies have proposed algorithms for automatic extraction and classification of lesions from leaf images. In this paper, the authors present a brief survey of the state-of-art and their contributions towards automatic recognition of disease lesions using deep learning methods. In particular, this paper highlights two deep learning models named KijaniNet and SwapNet which were proposed for use in automatic lesion extraction and classification algorithms. The paper concludes by suggesting some research points to be considered in future studies.