AgriNet-Adapt: A Hybrid Deep Learning Framework for Fruit Disease Detection Using ResNet-18 and EfficientNet-B0

Abstract

Accurate and scalable fruit disease identification remains a critical challenge in precision agriculture due to significant variations in disease appearance, environmental conditions, and image quality. This paper presents AgriNet-Adapt, a hybrid deep learning framework that integrates crop-specific optimized architectures for automated disease classification in pomegranate, mango, and guava datasets. The framework utilizes ResNet-18 and EfficientNet-B0 to achieve robust and high-precision performance across diverse fruit disease categories. The key novelty lies in an adaptive, model selection strategy termed Algorithmic Agronomy that replaces one-size-fits-all approaches by aligning deep learning architectures with crop characteristics, improving accuracy and efficiency.