Rice Varieties Identification using Deep Learning
Rice serves as a fundamental dietary staple for a vast majority of the global population. Given the diversity in grain types each possessing unique nutritional profiles and market values maintaining purity and authenticity in the supply chain is a critical challenge. Traditional manual sorting and identification processes are often slow and susceptible to human oversight. To modernize this workflow, we have developed a high-efficiency Deep Learning system capable of automatically distinguishing between various rice types using digital imagery.
Our solution is implemented using a robust Python-based Flask architecture, featuring a sleek web interface for real-time analysis. The core research evaluates two prominent neural network architectures: DenseNet-121 and the lightweight MobileNet. Utilizing a massive dataset of 60,000 high-resolution images, the system is trained to identify five distinct commercial varieties: Basmati, Jasmine, Arborio, Ipsala, and Karacadag.
While both models showed exceptional performance, MobileNet emerged as the superior choice for real-time applications, achieving a remarkable 99.5% test accuracy. By providing a fast and reliable method for grain classification, this project offers significant value for agricultural digitalization, helping stakeholders ensure food quality and prevent variety mixing in automated grain processing facilities.