The width of shredded tobacco is an important quality indicator in cigarette manufacturing, but reliable measurement is difficult because tobacco shreds are often fine, curled, fragmented, and densely overlapped. In this study, we propose a You-Only-Look-Once-Version-11-Nano (YOLO11n)-based visual selection and width measurement framework for identifying isolated and morphologically suitable tobacco shreds from complex production-line images. The proposed model integrates a bidirectional feature pyramid network, dynamic head, and scale-aware bounding box regression loss to improve multi-scale feature fusion, feature representation, and localization stability for slender targets. The images collected from an actual tobacco primary processing line were annotated according to measurement suitability and used for model evaluation. The improved model achieved a mean average precision (mAP) of 86.61% and an F1-score of 0.7935, representing improvements over the original YOLO11n of 11.15 percentage points in mAP and 0.0794 in F1-score, with an inference speed of 19.19 frames per second. The selected tobacco shred regions were further measured by a variable-radius circle method. Field validation on three production batches, covering 300 tobacco shreds, showed an average relative error of 4.66%, below the enterprise acceptance threshold of 8%. These results demonstrate that the proposed framework can effectively support tobacco shred width inspection under tested production-line conditions.