A two-stage method for defect detection and mechanical damage assessment in mechanically harvested tobacco leaves

Defects on mechanically harvested tobacco leaves may originate from diseases and pests or from harvesting operations, with distinct implications for crop quality. Dense small defects are easily missed, whereas local detection boxes cannot describe whole-leaf structural damage. We developed a two-stage vision method for non-destructive defect screening and assessment of harvest-related mechanical damage. Stage I used Tobacco Leaf Defect YOLO11n (TLD-YOLO11n), a task-adapted detector with high-resolution feature fusion, to identify four disease and pest categories and mechanical damage. Stage II used reference-midrib and leaf-morphology features to distinguish midrib breakage from lamina loss, while a relative lamina loss index provided an operational measure of tissue-loss severity. TLD-YOLO11n achieved mAP_0.5 and mAP_0.5:0.95 values of 84.5% and 64.1%, exceeding YOLO11n by 4.0 and 3.9 percentage points, respectively. Under controlled single-leaf imaging conditions, all 50 images with mechanical damage were passed to stage II, whereas none of the 156 disease- or pest-affected images were incorrectly passed. Damage type was correctly determined for 47 of the 50 mechanical-damage test images, yielding a joint accuracy of 94.0%. F1-scores for lamina loss and midrib breakage were 91.4% and 95.4%, respectively. The framework provides an approach for distinguishing harvest-related damage from other visual defects and quantifying loss of structural integrity in mechanically harvested tobacco leaves.