An Improved Deep Neural-Based Approach for Classifying and Identifying Plant

Abstract

The study presents a novel classification method designed to address the challenges posed by the identification of rice diseases, ultimately contributing to the enhancement of both crop yield and quality. With a multitude of rice diseases-over two dozen in number-posing significant threats to rice crops, the conventional manual diagnostic procedures have proven to be intricate and labor-intensive. In response, this research proposes the application of an automatic disease classification system, harnessing the power of deep learning-based image analysis techniques. This study specifically concentrates on the analysis of two key components of the rice plant-namely, the spikes and leaves-parts that have been shown to be the most vulnerable to disease manifestation. The method capitalizes on the unique characteristics exhibited by these parts of the plant to discern and categorize common diseases that afflict rice crops. The proposed approach achieves a remarkable testing accuracy rate of 98.38%, showcasing its proficiency in accurately distinguishing between different disease types. In terms of performance, the proposed classification methodology surpasses the capabilities of well-established deep learning architectures such as Visual Geometry Group 16 (VGG16) and Residual Network (ResNet50). It boasts a superiority of 7.51 and 16.22% in terms of performance accuracy metrics, respectively, thus underlining its robustness and efficacy. Additionally, the assessment of multiple performance metrics, including precision, sensitivity, and f-score, corroborates the exceptional performance demonstrated by the proposed approach.

Keywords: deep learning, crop health, VGG16, ResNet50, rice disease classification