Lightweight Hybrid Deep Learning Model for Real-Time Plant Disease Classification

Abstract

Plant diseases remain one of the leading causes of agricultural yield loss, food insecurity, and economic damage worldwide. Early and accurate disease identification is essential for minimizing crop losses and improving agricultural productivity. Traditional disease diagnosis relies heavily on manual field inspection performed by agricultural experts, which is often time-consuming, labor-intensive, subjective, and impractical for large-scale farming environments. The rapid advancement of computer vision and deep learning has significantly improved automated plant disease recognition; however, many existing deep learning models require substantial computational resources, making them unsuitable for deployment on smartphones, drones, edge devices, and Internet of Things (IoT)-based agricultural systems. Consequently, there is an increasing need for lightweight yet highly accurate deep learning models capable of performing real-time plant disease classification under practical field conditions. This study proposes a Lightweight Hybrid Deep Learning Model (LHDL-PDC) for real-time plant disease classification by integrating efficient convolutional neural network architectures with hybrid feature-learning mechanisms. The proposed framework combines lightweight feature extraction, image preprocessing, transfer learning, feature fusion, intelligent classification, and real-time decision support into a unified architecture suitable for precision agriculture. The objective is to achieve high classification accuracy while minimizing computational complexity, inference time, memory consumption, and energy requirements for deployment on resource-constrained devices.