PotatoGuardNet: a refined deep learning framework for potato leaf disease detection.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Potato production faces significant threats from environmental changes and crop diseases, impacting yield and quality.
• Manual disease classification methods are time-consuming, require expertise, and struggle with subtle symptoms.
• Automated systems are crucial for accurate, rapid disease detection to mitigate yield losses.

Purpose of the Study:

• To propose an improved deep learning approach, PotatoGuardNet, for accurate localization and classification of potato leaf diseases.
• To address the challenges posed by complex environmental conditions and variations in disease presentation.
• To develop a reliable automated system for agricultural disease monitoring.

Main Methods:

• An Inception-ResNet-V2 based Faster R-CNN model (PotatoGuardNet) was developed.
• The InceptionResNet-V2 network served as the base for feature extraction.
• A two-stage Faster R-CNN detector was employed for disease recognition and classification.

Main Results:

• PotatoGuardNet achieved a classification accuracy of 99.41% on the PlantVillage dataset.
• The model reported a mean Average Precision (mAP) of 0.9556.
• Heatmaps were generated to demonstrate the model's explanatory power and localization capabilities.

Conclusions:

• PotatoGuardNet demonstrates high effectiveness and reliability in detecting and classifying potato leaf diseases.
• The model successfully captures disease-specific visual patterns, outperforming state-of-the-art approaches.
• The findings suggest PotatoGuardNet's potential for practical deployment in automated agricultural disease monitoring systems.