MBConvNet-CNS: A Lightweight Convolutional Neural Network (CNN) Architecture for Plant Disease Detection

Plant diseases pose a big threat to agricultural productivity worldwide, so their detection technique should be strong and efficient. Classically, plant disease identifications are time-consuming and usually require expert knowledge; this makes the use of large-scale agriculture difficult. The following paper presents a review of how the use of lightweight CNN models can be used for real-time plant disease detection with models that are computationally efficient enough to be deployed on edge devices such as smartphones and drones. We shall propose a novel model based on state-of-the-art architectures comprising the likes of MobileNet, EfficientNet, Ghost Modules, and Squeeze-and-Excitation blocks that achieve high accuracy for plant disease identification with a low computation footprint. The proposed model was applied to different datasets of images of plant leaves. Experimental results proved it at par to the previous heavyweight and lightweight models based on deep learning for accuracy and time consumed. These findings point out that the correct integration of lightweight CNN architectures and innovative modules shall yield a feasible solution for effective management of plant diseases, making timely interventions possible for reducing crop losses.