Role of Dimensionality Reduction Techniques for Plant Disease Prediction

Abstract

Plant diseases are one of the main reasons for the degradation of a plant’s productivity and quality, which can lead to overall economic loss. Therefore, plant disease management has become an important area of interest for agriculturists. The manual detection, prediction, and diagnosis of plant diseases is a challenging task that requires deep knowledge about plants and diseases that can affect them. The trend to use computers in agriculture for the improvement and safety of plants and trees is gaining popularity. Nowadays, numerous machine learning (ML) models and classifiers have been introduced and deployed for the prediction and classification of various plant leaf diseases. However, the high dimensionality of plant leaf images makes it difficult for plant disease prediction. To overcome this issue, dimensionality reduction techniques (DRTs) can play a significant role in reducing the dimensionality of data by extracting an abstract representation of high dimensional data to explore actionable insights. In this study, we used DRTs to reduce the dimensionality of plant images to proliferate the performance of ML models developed for the analysis and processing of plant images. The successful application of DRTs has dramatically enhanced the performance of ML models developed for the recognition, prediction, and diagnosis of diseases. This trend offers a new perspective to the agriculture domain to predict, diagnose, and classify diseases. This chapter aims to explore the utilization of DRTs to reduce the dimensionality of images used for plant disease prediction, diagnosis, and classification via ML models. Some thought-provoking applications of DRTs for agriculture and particularly for plant disease prediction have been explored in this chapter. Some major issues and challenges of DRTs and the computer-based prediction of plant diseases have also been identified as open research issues. Finally, we conclude the chapter and mention the directions for future work in this evolving domain.