Rice seed classification using machine learning and deep learning

Rice is a staple crop for Indonesian citizens. The need for rice increases along with population growth, and farmers need quality rice seeds. This study compares machine learning and deep learning to identify rice seeds. The method used is artificial intelligence by applying machine learning and convolutional neural networks. The shape feature extraction approach is necessary for machine learning. Convolutional neural networks use a group of labeled image data to train. The author's contribution is to use 34 custom layers to reduce the computing burden of deep learning. The secondary data used are five rice seeds classes, Arborio, Basmati, Ipsala, Jasmine, and Karacadag, with 1000 files per class. A combination of methods is needed to obtain numerical and image validation. The test results for the two classifications of rice seeds show the accuracy of machine learning with random forests and extreme learning machines at 98% with an average time of 2 minutes. Convolutional neural networks with 34-layers obtained an average accuracy of 99.92% and an average training process computing time of 20 minutes and 54 seconds. However, the testing process only takes a few seconds.