Lightweight convolutional neural network-based plant disease identification for protection and landscape design

Abstract

Plant diseases significantly impact landscape design, necessitating comprehensive consideration and effective management measures to ensure the health, aesthetics, and sustainability of landscapes. Early detection and timely control of plant diseases are crucial, but traditional monitoring methods, which rely on manual observation and sample collection, are inadequate for covering large garden areas and may delay necessary treatments. This study addresses these challenges by constructing a small Rosa chinensis disease dataset through field collection and data augmentation techniques. We propose MixResCoAtNet, an improved model based on the lightweight MixNet framework, to identify and categorize diseases from plant leaf images using convolutional neural networks (CNNs). Comparison experiments with various classical convolutional network models demonstrate that MixResCoAtNet outperforms these models, offering more competitive performance. Additionally, due to its lighter structure, MixResCoAtNet shows greater potential for deployment on mobile devices, facilitating real-time and efficient plant disease monitoring and management in landscape design.