An Improved RT-DETR Model for Detection of Tobacco Impurities

Tobacco, as a crop of significant economic importance, plays a pivotal role in global agriculture and industry. However, during the processing of tobacco products, the presence of tobacco impurities (TIs) poses a major challenge, as it substantially diminishes product quality. Traditional methods for detecting TIs are often inefficient and limited in scope, underscoring the need for precise and reliable detection techniques. This study addresses these challenges by introducing an improved Real-Time Detection Transformer (RT-DETR) model. By integrating the ShuffleNet V2 architecture and a novel Multi-Branch-Edge-Information-Select (MBEIS) block, the model enhances feature extraction and fusion capabilities, enabling efficient detection even in complex scenarios. Furthermore, a standardized imaging environment, consisting of controlled lighting and a black box setup, was employed to ensure consistent data collection, and the dataset was augmented using advanced data enhancement techniques. Experimental evaluations demonstrate that the proposed model achieves a precision of 92.7%, a recall of 93.5 %, and a mean average precision (mAP) of 87.9%, outperforming mainstream detection algorithms. In addition, the reduction in model parameters and computational complexity ensures superior real-time performance, making it both lightweight and effective for industrial applications. The results highlight the model's strong potential as a robust solution for detection of TIs, providing a significant technological advancement for the tobacco industry. This research not only addresses critical quality assurance challenges but also offers a valuable reference for applying advanced detection technologies in other industrial domains.