Towards Precision Agriculture: Deep Learning-Based Plant Disease Identification

Abstract

Existing plant disease identification models encounter several limitations, primarily due to their inability to effectively capture the complex relationships within multifaceted plant images. Traditional deep learning methods often struggle to accurately detect diseases across various plant components, resulting in suboptimal performance. Additionally, most existing models focus on specific plant parts, such as leaves, rather than considering multiple components—such as leaves, stems, fruits, and roots for a more comprehensive diagnosis. Another major limitation is the absence of dynamic parameter optimization in conventional models, which reduces their adaptability for real-time applications. Furthermore, current approaches do not effectively account for the continuous evolution of plant diseases, making them less reliable for practical agricultural use.

To overcome these challenges, this research introduces two advanced deep learning-based models: the Multivariate Normal Deep Learning Neural Network (MNDLNN). These models enhance disease identification accuracy, sensitivity, and specificity by leveraging sophisticated learning techniques. The MNDLNN classifier improves disease detection by analysing multiple plant components, ensuring a more holistic approach. Meanwhile, the MPO Adapted Deep NN enhances precision through continuous learning capabilities, enabling more accurate disease identification. A key innovation of this study is the integration of the Modified Political Optimization (MPO) algorithm, which dynamically fine-tunes classifier parameters to optimize performance. These findings mark a significant advancement in plant pathology, providing a highly accurate and adaptable framework for real-time agricultural disease monitoring.