Abstract

Rice leaf diseases plays a critical role in agricultural productivity and food security, particularly in countries with high rice consumption, such as Indonesia. This study proposes a lightweight classification model that integrates Convolutional Autoencoder (CAE) and Convolutional Neural Network (CNN) to classify five types of rice leaf diseases. Unlike previous studies that rely on large pre-trained models, this study focuses on building a custom architecture from scratch, emphasizing model efficiency and simplicity. CAE is used to extract 128-dimensional latent features from input images, which are then classified using a shallow CNN architecture. Global Max Pooling is introduced at the bottleneck stage to reduce the number of parameters without sacrificing performance. The model is trained and evaluated on a dataset containing 7,430 labeled samples and achieves 99.33% accuracy with only 0.41 million parameters. Comparing the suggested model to existing model reveals that the proposed model provides an optimal balance between model compactness and predictive performance, making it ideal for future low-resource deployments.

Concepts :

Access to Document

SDGs

Citations by Year