Comparative Analysis of Deep Learning Architectures for Rice Crop Image Classification

Rice, a vital grain globally has various varieties around the globe. Classifying these varieties requires usually requires an experienced eye. However, by leveraging Deep Learning models, this task can be made significantly easier. This research paper explores the subtleties of five distinct rice varieties using five different Deep Learning models: ResNet50, ResNet50v2, VGG16, ConvNextTiny, and DenseNet169. The study delves into analysing the performance of these models which utilise Transfer Learning on the task of rice-grain classification. The selected models undergo a meticulous evaluation, following contemporary research procedures and assessing criteria such as accuracy, precision, recall, F1 score, and AUCROC score. Leveraging the extensive Rice image collection featuring 75,000 images, the research culminates in an outstanding 99.64% accuracy in rice classification. A high accuracy for all models deployed helped put confidence into the idea that different approaches can be undertaken towards the rice classification task based on the requirements and resources available. Positioned at the intersection of technology and agriculture, this research paper serves as a guiding beacon, showcasing the transformative potential of Deep Learning models in overcoming persistent challenges in rice classification assessment. Positioned at the intersection of technology and agriculture, this research paper showcases the transformative potential of Deep Learning models in overcoming persistent challenges in agriculture.