Developing an eAgriculture application for identification of fungal disease in plants through leaf images

Abstract

Development of automatic disease detection and classification system is significantly explored in precision agriculture. In the past few decades, researchers have studied several cultures exploiting different parts of a plant. The symptoms of plant diseases are evident in any part of a plant, however leaves are found to be the most commonly observed one for infection identification. Researchers have thus attempted to automate the process of plant disease detection and classification using leaf images. Several works utilized computer vision technologies effectively and contributed a lot in this domain. The work presented in this study summarizes the pros and cons of all such studies to throw light on various important research aspects. A discussion on commonly studied infections and research scenario in different phases of a disease detection system is presented. The performance of state-of-the-art techniques are analyzed to identify those that seem to work well across several crops or crop categories. Discovering a set of acceptable techniques, the study presents a discussion on several points of consideration along with the future research directions. Based on the understandings gained during the survey, a computer vision based systems for plant disease detection using leaf images are developed. The main culture focused in this study is soybean due to its several benefits. A rule-based system using concepts of kmeans is designed and implemented to distinguish healthy leaves from diseased leaves. The system works with a set of rules proposed in this study. Once a leaf is identified as unhealthy, it is classified into one of the three categories (downy mildew, frog eye, and septoria leaf blight) effectively by utilizing the framed rules. The efficacy of the system is proved by performing experiments separately on various color features, texture features and their combinations. Results are generated using thousands of images collected from PlantVillage dataset. Acceptable average accuracy values are reported for all the considered combinations which are also found to be better than existing ones. An attempt has also been made to discover the best performing feature set for leaf disease detection in soybean. The system is shown to efficiently compute the disease severity as well. Qualitative as well as quantitative measures are utilized to further prove suitability of the proposed system in detection, classification, and severity calculation. Another area focused is to design a generalized framework that can detect a leaf image as healthy or unhealthy and in case of diseased identify its type. The basic idea in developing this system is to eliminate the rules completely. It is a two-stage framework and termed as semi-automatic system in this work. It continues with the same concepts as is utilized in rule-based system in addition to the concepts of two-/multi-class classifiers. Classifiers are trained on numerous features (texture and color). The framework may employ fusion technique in case of more than one classifier and is flexible to work with the best performing classifiers. The study discusses a fusion method as well to combine the results logically. The framework, i.e. semi-automatic system is tested on six different datasets formed with leaf images from legumes, vegetables, fruits, and commercial crops to verify the notion of generalization. Satisfying results are achieved in each of the considered cases. Both the developed systems are compared with several existing systems in literature and are found to perform better on the following parts: image acquisition, segmentation, and number of training/testing images. The systems have shown agreeable performance on three cross-domain scenarios too. For better comparison, three papers are implemented and tested on PlantVillage dataset, here too the proposed systems outperforms. Lastly, on generalization characteristic of the semi-automatic systems results are good considering the system simplicity. However, there exist a few deep-learning based systems which are superior but in the absence of any standardized datasets the results presented here are acceptable. A web-application is also developed with the proposed semi-automatic system that serves as a good assistance to any naïve user too.