Rice Grain Habitat Identification System using Convolution Neural Network on Hyperspectral Imaging

Indonesia is one of the world's nations, specifically in the Asian continent, that consumed rice as their main dishes. This phenomenon led rice to be highly demanding food and planted in many regions. However, the habitat rice plantation contributes to the rice quality, taste, and fragrance that impact the particular rice market price. This research proposed an identification system to differentiate the rice grain properties based on the rice grain's spectral and spatial information. The proposed identification system consists of the workbench and the algorithm that computed hyprcube data from the camera. The hyperspectral camera in the range 400-1000 records the rice grain sample placed in the petri dish. The image correction and segmentation were performed t enerate the 3D image rice sample as an input of the Convolution Neural Network (CNN). The CNN was modified from the autoencoder approach to classifying the rice habitat. Five types of rice habitat from different planting areas were used to measure the system performance, such as Bandung, Indramayu, Subang, Karawang, and Palembang. A total of 480 rice data sets were used to compute the accuracy of the CNN classification model. The accuracy of CNN is 100% at the training and 94% at the test.