PJB-2024-108
POA based Ontology-Guided Potato Leaf Disease Detection Using NAM-PSPNet Segmentation and ENet-TAM Classification
V.Rajeswari
Abstract
A primary objective of this research is developing an advanced automated technique for the early identification and categorization of potato leaf diseases (PLD), one of the most extensively farmed vegetable crops worldwide. The number and quality of potatoes produced worldwide are significantly impacted by these illnesses, which include the early and late blights brought on by Alternaria solani and Phytophthora infestans. This paper offers a comprehensive technique to detect potato leaf disease leveraging the PlantVillage dataset. The preprocessing stage incorporates min-max normalization to effectively remove noise and enhance image quality. Subsequently, an ontology-based feature selection strategy is employed using the Pufferfish Optimization Algorithm (POA) to identify salient disease-related features. This study implemented a novel approach to feature selection utilizing POA. POA efficiently identifies crucial features in an ontology-driven manner, seamlessly integrating them into the segmentation pipeline for enhanced performance and precision. To achieve accurate delineation of diseased areas, an enhanced Pyramid Scene Parsing Network (PSPNet) that incorporates the Attention Module based on Normalisation (NAM) is employed for segmentation called as NAM-PSPNet. Furthermore, for end-to-end classification, an EfficientNet model is enhanced with a triplet attention mechanism (ENet-TAM) to accurately categorize various potato leaf diseases. The proposed ENet-TAM model showcases exceptional performance with an accuracy of 98.62%. Its precision, sensitivity, and F1-score values all exceed 98.1%, highlighting its robustness in classification tasks. These outcomes highlight how successful the model in accurately categorizing data instances, making it a viable option for a range of practical uses.
To Cite this article:
Download
