A Lightweight Convolutional Neural Network for Rice Diseases Classification

• Published in: 2024 IEEE 14th International Conference on Control System, Computing and Engineering (ICCSCE) pp. 214 - 219
• Main Authors: Ismail, Mohd Aniq, Rashid, Mohamad Rahmat, Ahmad, Khairul Azman, Setumin, Samsul, Bakar, Siti Juliana Abu, Ani, Adi Izhar Che
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23-08-2024
• Subjects: Accuracy; classification task; CNN; Computer architecture; Convolutional neural networks; Crops; deep learning; Diseases; Image edge detection; Image enhancement; Image segmentation; K-means; lightweight; Nutrients; rice disease; Training; VGG16
• DOI: 10.1109/ICCSCE61582.2024.10696617

Rice, also known as Oryza Sativa, is widely recognized as an important source of food crops and a primary source of nutrients for human beings as it is rich in carbohydrates, providing a high-energy food. The United Nations General Assembly stated that rice is the staple food of more than half the world's population. However, pests and disease outbreaks are the major problems for farmers as this rice crop is susceptible to fungi and viruses. Leaf Brown Spot, Leaf Blast, and Leaf Hispa are among the most well-known diseases that affect the rice crop. Accurate and fast classification of these diseases is essential for implementing timely and targeted approaches to mitigate crop losses. This paper investigates the feasibility of our proposed lightweight model to be used for the classification task with a limited number of training samples. The images were preprocessed to enhance the color and edges. The K-means clustering method was used for image segmentation to separate the affected regions from the background. The proposed convolutional neural network model was made from scratch to have only three convolutional layers. The results demonstrate that the proposed model outperforms the VGG-16 model for overall performance with 90% and 85% for both batch sizes.