IEEE Access (Jan 2021)

A Generic Approach for Wheat Disease Classification and Verification Using Expert Opinion for Knowledge-Based Decisions

  • Waleej Haider,
  • Aqeel-Ur Rehman,
  • Nouman M. Durrani,
  • Sadiq Ur Rehman

DOI
https://doi.org/10.1109/ACCESS.2021.3058582
Journal volume & issue
Vol. 9
pp. 31104 – 31129

Abstract

Read online

Crop diseases have mainly affected crop production due to the lack of modern approaches for disease identification. For many years, farmers have identified various crop diseases and have local knowledge about disease management. However, the local knowledge of one agricultural region is not utilized in other regions due to the unavailability of knowledge sharing platforms. Agricultural research also suggests that crop production has mainly decreased due to diseases, methods of cultivation, irrigation, and lack of local agricultural knowledge. In this research, the experience of agricultural experts, farmers, and cultivators is gathered through a crowd-sourced platform. The data is then processed for various disease identification. Hence, timely identification of various crop diseases can benefit farmers to apply relevant management methods. In literature, researchers have proposed various methods for disease management, mostly based on the classification of crop diseases using Machine Learning (ML) algorithms. However, these algorithms are unable to give trustful results due to static data provisioning and the dynamic nature of various diseases in different agricultural regions. Further, the agricultural expert's experience is also not considered in verifying the classification results. To identify the dynamic nature of wheat diseases, we acquired high-quality images and symptoms-based text data from farmers, domain experts, and users using a crowd-sourced platform. Different augmentation techniques were also used to enhance the size of training data. In this paper, a modern generic approach has been proposed for the identification and classification of wheat diseases using Decision Trees (DT) and different deep learning models. Also, results of both algorithms were then verified by domain experts that improved the decision trees accuracy by 28.5%, CNN accuracy by 4.3% (leading to 97.2%), and resulted in decision rules for wheat diseases in a knowledge-based system.

Keywords