Plant disease identification using automated image analysis

Abstract

Plant health is of the universal importance as cultivated plants are a primary source of energy for human life. To maintain quality and quantity of crop productivity, early disease diagnosis is required. Providing an expert for in-field continuous crop monitoring can be costly and time-consuming, especially in remote areas. An automated system for plant disease identification can be developed using image processing algorithms to recognise visible symptoms of plant diseases. Development of an automated system for use in practice is a challenging task, and existing systems developed for agricultural applications are still in their early stages. Factors affecting performance of existing systems include the presence of background content in images and inconsistency of images captured under unconstrained conditions. Thus, there exist opportunities for developing a practical imaging framework for characterising plant diseases. Our proposed system framework is based on fundamental steps in image processing, including pre-processing, segmentation, feature extraction and classification. Pre-processing and segmentation techniques are developed to handle image inconsistency and to remove background content, which is unavoidable in practical applications. Investigation of various features and the system contribution of individual features including texture, colour features and region properties, is performed. In addition, feature representations are introduced to describe natural plant disease patterns and to cope with different levels of disease severity. An evaluation of the proposed system indicates that the pre-processing and segmentation algorithms applied in the system improve overall classification accuracy significantly. A combination of different feature sets, which are potentially pre-selected, provides the most accurate classification of disease leaf images acquired under both controlled and uncontrolled conditions. A prototype of the imaging framework integrated with mobilecloud computing is also investigated, and the integrated system demonstrates improvement in reliability and mobility suitable for a practical situation.

Date of Award	Oct 2016
Original language	English
Sponsors	Vice Chancellor's Research Scholarship (VCRS) & Partial support from EPSRC funded India-UK Advanced Technologies Centre project
Supervisor	Bryan Scotney (Supervisor), Philip Morrow (Supervisor) & Gerard Parr (Supervisor)