Authors

Abstract

As a vital food crop, rice plays a crucial role in the global food supply. Accurate seed sorting is critical for planting and sales, but traditional variety identification methods are time-consuming, inefficient, and prone to causing physical damage to seeds. To enhance identification efficiency and classification accuracy, this study employed an image acquisition system to capture images of eight locally grown rice seed varieties. After preprocessing and segmenting the original images to improve data quality, multi-dimensional features were extracted and analyzed to construct a deep learning model for rice seed identification. The results showed that the Rice-Transformer model, based on the Transformer architecture, achieved a classification accuracy of 97.71%, demonstrating excellent identification capabilities. Additionally, this study developed a user interface based on PyQT5 to visualize the identification results. It can provide a feasible solution for the efficient and non-destructive identification of rice seed varieties and has the potential to be applied in consumer markets and the food industry.

Abstract in Chinese

水稻作为重要的粮食作物，在全球粮食供应中占据关键地位。种子的精准分选对其种植与销售环节至关重要，然而传统的品种鉴别方法存在耗时、低效且易对种子造成物理损伤的问题。为提高鉴别效率和分类准确性，本研究通过图像采集系统获取了八种本地主要种植的水稻种子图像。在对原始图像进行预处理与分割以提升数据质量的基础上，提取其多维度特征进行分析，构建了用于水稻种子鉴别的深度学习模型。结果表明，基于Transformer架构的Rice-Transformer模型分类准确率高达 97.71%，表现出优异的鉴别能力。此外，本研究基于PYQT5开发了用户交互界面，实现了鉴别结果的可视化展示。该研究可为水稻种子品种的高效、无损鉴别提供可行方案，具备应用于消费市场与食品工业的潜力。