Disease Detection & Severity Prediction of Pomegranate Fruit

EasyChair Preprint 14222

Abstract

This paper presents an integrated framework for the comprehensive analysis of diseases affecting pomegranate fruit. Leveraging deep learning techniques, the proposed framework encompasses multiclass disease classification, semantic segmen- tation for diseased area localization, and severity estimation through image processing. Initially, a Convolutional Neural Net- work (CNN) is employed for multiclass classification, effectively identifying four prevalent diseases: Alternaria, Anthracnose, Bacterial Blight, and Cercospora. Subsequently, Semantic Seg- mentation using the UNet architecture is utilized to delineate the diseased areas within the fruit, requiring annotated images as training data. The segmentation results serve as inputs for severity estimation, achieved by calculating the area and percent- age of the segmented region, followed by defining thresholds for severity levels (high, medium, low). The severity level prediction is based on the percentage of the segmented region, facilitating early detection and intervention strategies. The proposed framework offers a comprehensive approach to disease analysis, integrating classification, semantic segmentation, and severity prediction to enhance pomegranate fruit management practices.

Keyphrases: Convolutional Neural Networks, UNet, annotation, deep learning, semantic segmentation

@booklet{EasyChair:14222,
  author    = {Poulin Paul and S Manishankar and R. Satheesh Kumar},
  title     = {Disease Detection & Severity Prediction of Pomegranate Fruit},
  howpublished = {EasyChair Preprint 14222},
  year      = {EasyChair, 2024}}