CROP DISEASE DETECTION USING MACHINE LEARNING

Abstract

Humans rely on cereals as a major food source, emphasizing the need for increased crop production to sustain the growing population. However, plant diseases significantly impact crop yield and food quality. Wheat, a crucial crop worldwide, is particularly vulnerable to diseases. Early detection and classification of these diseases are vital for effective disease management. In this study, we conducted a comprehensive literature analysis of wheat diseases from 2017 to 2022. We identified three major types of wheat diseases, namely fungal, bacterial, and insect related. Additionally, we reviewed 32 studies focused on disease detection and classification using various machine learning and deep learning algorithms. Our analysis revealed that Stripe Rust garnered the most attention, accounting for 56% of the total studies. Self-acquired datasets were predominantly used, and convolutional neural networks (CNN) and its frameworks were the most prevalent classification techniques, representing 34% of the studies. Accuracy emerged as the dominant performance metric, constituting 65% of the total studies. Notably, the majority of the literature was published in 2019 (25%) and 2020 (25%). Considering the urgency of crop health and productivity, we propose a hybrid model that combines the strengths of Convolutional Transformer and EfficientNet architectures for reliable wheat disease classification. Our model integrates global contextual information gathering from Convolutional Transformer with the efficiency and accuracy improvements of EfficientNet. To train the model, we preprocess and augment a dataset comprising 14,560 images of Fusarium head blight, Yellow rust, Brown rust, Powdery mildew, and healthy wheat leaves. With an impressive accuracy of 93.6%, our proposed model offers valuable insights for agricultural disease management, enabling enhanced crop health monitoring and ultimately improving productivity and sustainability. This research contributes to addressing the challenges associated with wheat disease detection and classification, paving the way for more efficient agricultural practices.