Precise measurement of the thousand-grain weight is critical for accurately forecasting rice yields. This parameter is essential for variety development and appropriate cultivation management. Precise identification and enumeration of rice grains are required for accurate measurements of thousand-grain weight, a pivotal stage in the research process. However, the procedure has substantial challenges due to the small size of rice grains, their intrinsic similarity, and the differing degrees of stickiness. The TCLE-YOLO model is an advanced deep learning technique that integrates a transformer encoder and coordinate attention module. It utilises the robust YOLOv5 as the underlying network architecture. The model integrates a coordinate attention (CA) module into the YOLOv5 backbone to enhance feature representation in small target regions. The system also incorporates a specialist detection head designed for detecting small targets. This head utilises a feature map with high resolution and low-level details. Furthermore, the neck module employs a transformer encoder to augment the network's capacity to include a broader spectrum of data and amplify the extraction of crucial characteristics from recognised targets. This increases the sensitivity of the additional detecting head specifically towards rice grains, particularly those that have a significant level of adhesion. The implementation of EIoU loss significantly improves accuracy. The results of our experiments on our custom rice grain dataset demonstrate outstanding precision, recall, and mAP@0.5 scores of 99.20%, 99.10%, and 99.20%, respectively, surpassing several state-of-the-art models. The TCLE-YOLO model we have developed offers a robust foundation for accurately identifying and quantifying rice grains. It provides vital information for accurate measurements of thousand-grain weight and the efficient assessment of rice breeding procedures.

Index Terms- About four key words or phrases in alphabetical order, separated by commas. Keywords are used to retrieve documents in an information system such as an online journal or a search engine. (Mention 4-5 keywords)