Analisis Perbandingan Arsitektur Deep Learning Pada Klasifikasi Penyakit Tanaman Berbasis Citra Daun: Systematic Literature Review

Abstract

Plant disease classification based on leaf images has become an important approach to supporting precision agriculture and early plant disease detection systems. This study aims to examine the development of Deep Learning methods for plant disease classification through a Systematic Literature Review (SLR) approach. The review analyzed 15 scientific articles covering various crops, including rice, maize, tomato, tea, cassava, citrus, tobacco, and coffee. The literature selection process followed the PRISMA framework. The analysis focused on comparing Deep Learning architectures, dataset characteristics, and model performance based on accuracy metrics. The findings indicate that conventional Convolutional Neural Networks (CNNs), modern architectures such as ResNet, Inception, EfficientNet, and DenseNet, as well as lightweight models such as MobileNetV2, are the most widely adopted methods. Model performance is influenced by dataset size and source, with large-scale public datasets generally achieving higher and more consistent accuracy. This study highlights that selecting an appropriate Deep Learning architecture requires balancing model complexity, data availability, computational efficiency, and practical deployment requirements to develop effective and reliable plant disease classification systems.