Comparative Evaluation of Machine Learning Models for Rice Seed Quality Identification on an Android Platform

The challenge of significantly increasing rice production in Indonesia necessitates a granular focus on the farm level, with particular emphasis on the utilization of high-quality seeds. Unfortunately, the accurate assessment of seed quality is often only feasible post-planting. While traditional methods persist, they are notably time-consuming and prone to inaccuracies. This research introduces a novel solution: an Android-based machine learning model for rice seed quality identification based on morphological seed structure. By enabling early detection of low-quality seeds, the system aims to mitigate the associated risks and contribute to increased rice yield. Two cutting-edge machine learning models were trained using datasets sourced from the Roboflow platform and subsequently integrated into an Android application. A comparative analysis of these models was conducted to determine their efficacy in discerning rice seed quality. The evaluation results unequivocally demonstrated the superior performance of the Roboflow Train 3.0 Object Detection (Fast) model, achieving an impressive mean average precision (mAP) of 97.1%, precision of 96.2%, and recall of 93.4%. Given its exceptional accuracy and speed, the Roboflow Train 3.0 Object Detection (Fast) model has been proven to be the ideal option for integration into Android applications. Future research could concentrate on optimizing the model for edge inference, thereby enhancing identification efficiency and accuracy. This advancement holds the potential to revolutionize rice cultivation practices in Indonesia by empowering farmers to make informed decisions based on precise seed quality assessment, ultimately leading to substantial improvements in rice production and food security.