Abstract: To enhance efficiency and accuracy of germination detection for ultra-small seed and meet demand for automation, shepherd's purse seed were selected as research subject．Through shepherd's purse seed germination experiments' image analysis, a YOLOv7-tiny-based seed germination detection and recognition method was improved and designed to achieve rapid detection of shepherd's purse seed germination, and detection experiments were conducted．Experimental results showed that YOLOv7-tiny achieved a germination discrimination accuracy of 92.7% for shepherd's purse seed, while improved DW-YOLOv7-tiny reached 96.2%．Compared to original data, parameter count and computational load were decreased by 3.32% and 4.35%, respectively．Model's accuracy, recall, and average precision mean increased by 3.5, 2.5, and 1.8 percentage points, respectively．Model's weight file size was reduced by 1.0 MB, achieving greater lightweight performance while maintaining a detection speed of 121 f/s．These results demonstrate effectiveness of improved method, which is particularly suitable for rapid germination detection of ultra-small seeds like shepherd's purse in specialized application scenarios．