Performance Evaluation of Rice Seed Classification System Based on The CNN with VGG-GoogleNet Architecture

The success of the plant and its ability to continue to produce for a long time is dependent on the seed. Balai Pengawasan Sertifikasi Benih Tanaman Pangan dan Hortikultura (BPSBTPH) West Java Province is entrusted selecting high quality seeds through a certification program as one of the stakeholders in Indonesia’s agricultural ecosystem. The certification process is still handled by the conventional strategy, relying entirely on the observations and knowledge of qualified assessment professionals who are experts in their professions. This research creates and evaluates rice seed classification models based on the CNN with VGG-GoogleNet architecture. The VGG-GoogleNet model is created by utilizing two types of datasets taken exclusively from the BPSBTPH laboratory, which are rice seed growth dataset and rice seed quality dataset. The evaluation shows that the accuracy of the VGG-GoogleNet model exceeds other models, which are LeNet, AlexNet, ZfNet, VGGNet, and GoogleNet. The accuracy of the VGG-GoogleNet achieves 98.25% and 94.99% for the rice seed growth dataset and the rice seed quality dataset respectively.