Budi Prasetiyo, Fadhl Al-Hafizh, M. Faris Al Hakim, Sri Sukaesih
Rice (Oryza sativa) is a vital global food crop, despite its importance, rice production is often hindered by leaf diseases, such as Leaf Scald and Bacterial Leaf Blight, which can significantly reduce yields. Conventional disease diagnosis techniques are often error-prone and inefficient. To address this, A deep learning-based approach using EfficientNetB4 architecture was proposed and combined with a Gaussian filter for enhanced image preprocessing. The Gaussian filter reduces noise and enhancing image, while EfficientNetB4 leverages its optimized depth, width, and resolution scaling for accurate classification. The dataset consists of 2,627 rice leaf images categorized into six classes and divided into training, validation, and testing. Preprocessing includes resizing images to 224 × 224 pixels, data augmentation, and Gaussian filtering with 5 × 5 kernel and standard deviation value is 1. The model is evaluated using F1-score, precision, recall, and accuracy. Results demonstrate that EfficientNetB4 with Gaussian filtering achieves 97.62% accuracy, outperforming the unfiltered model 96.19 %. This highlights the efficacy of Gaussian filtering in improving feature extraction and classification performance for rice leaf diseases. © 2026 IEEE.
Universitas Negeri Semarang, Department of Computer Science, Indonesia; Universitas Negeri Semarang, Department of Biology, Indonesia
