An Efficient Classification of Rice Variety with Quantized Neural Networks

Rice, as one of the significant grain products across world, features a wide range varieties in terms usability and efficiency. It may be known with various regional names depending on specific locations. To specify particular rice type, different are considered, such shape color. This study uses an available dataset Turkey consisting five varieties: Ipsala, Arborio, Basmati, Jasmine, Karacadag. The introduces 75,000 images total; each 5 has 15,000 samples 256 × 256-pixel dimension. main contribution this paper is to create Quantized Neural Network (QNN) models efficiently classify purpose reducing resource usage edge devices. well-known that QNN successful method for alleviating high computational costs power requirements response many Deep Learning (DL) algorithms. These advantages quantization process have potential provide efficient environment artificial intelligence applications microcontroller-driven For purpose, we created eight networks using MLP Lenet-5-based deep learning varying levels trained by dataset. With network at W3A3 level, 99.87% classification accuracy level was achieved only 23.1 Kb memory size used parameters. In addition tremendous benefit usage, number billion transactions per second (GOPs) 23 times less than similar studies.