Abstract: To address the issue of low detection efficiency and insufficient removing accuracy caused by the small size, similar color to tobacco leaves and complex background interference of non-tobacco related materials (NTRMs) from tobacco leaves, a NTRM removing system based on the collaboration of improved YOLOv10 algorithm and robot hands was designed. Based on the original YOLOv10 algorithm, the improved convolutional block attention module (CBAM) was introduced and small-size object detection heads were used to achieve fast recognition and precise localization of NTRMs. Eight NTRMs with high occurrence frequencies in the production line were selected for analysis. The results showed that: 1) The improved YOLOv10 algorithm achieved an average precision of 94.8%, a recall rate of 97.4% and an elapsed time of 4.2 ms for NTRMs detection, significantly outperforming other models. Compared with the original algorithm, the improved algorithm increased the average precision by 1.1 percentage points, raised the recall rate by 0.9 percentage point and reduced the elapsed time by 0.3 ms. 2) The recognition accuracy of the removing system by robotic hands for 8 NTRMs was 92.6%, with a removal rate of 96.0%, effectively improving the quality of tobacco processing. This study provides a technical solution for intelligent quality control in tobacco leaf sorting for cigarette production.