Rice Leaf Disease Detection: A Comparative St...

Rice Leaf Disease Detection: A Comparative Study between CNN, Transformer and Non-Neural Network Architectures

Author Name : Jyoti Dinkar Bhosale, Priti Vijaykumar Panchol, Suraj Shankarrao Damre, Bharat Madhavrao Pawar

DOI: https://doi.org/10.56025/IJARESM.2025.1302250914

ABSTRACT A sizable section of the population in countries like Bangladesh depends on agriculture for their livelihoods. Early detection and classification of plant diseases is essential to stop their spread and lessen their negative effects on crop quality and output. For such identification and classification, a variety of computer vision algorithms may be used. Although CNNs have dominated these picture classification tasks, vision transformers have recently improved to an equivalent level. This work examines the different computer vision methods for detecting rice leaf disease in Bangladesh. We test several CNN and ViT models using the Dhan-Shomadhan dataset, which is a dataset of rice leaf diseases in Bangladesh. Additionally, we evaluated how well such deep neural network architectures performed in comparison to more conventional machine learning architectures, such as Support Vector Machines (SVM). With less training data, we improved generalization by using transfer learning. ResNet50 was the best option for this job out of all the models examined since it performed better than other CNN and transformer-based models.