Abstract

The major challenge in Sunn hemp seed production is contamination with seeds from other species that share similar physical characteristics, such as morning glory seeds. This research aimed to 1) explore the artificial intelligence application for analyzing Sunn hemp seed images and 2) evaluate the physical purity of Sunn hemp seeds. The results showed that the pre-trained Inception V3 model, using the Adaptive Moment Estimation (ADAM) algorithm, achieved the highest accuracy of 98.10% compared to other models like VGG19 and GoogLeNet (ImageNet), which achieved 97.37% and 98.00% accuracy, respectively (using the ADAM algorithm as well). These findings demonstrate that Inception V3 with ADAM is highly effective in distinguishing Sunn hemp seeds from morning glory seeds. This may be attributed to Inception V3's architecture, which is well-suited for complex image processing. Therefore, it has the potential to be applied for assessing seed physical purity on mobile phones and production lines.