Table 2.

Results of the ablation experiments

Group	Image preprocessing	ConvFFN	Nonlocal	Swin transformer	Precision (%)	Recall (%)	mAP@0.5 (%)	Detection time (ms)
1	×	×	×	×	85.4	75.8	86.0	20.1
2	√	×	×	×	91.9	78.1	88.8	20.1
3	√	√	×	×	91.3	83.3	91.3	20.7
4	√	√	√	×	92.6	83.0	91.9	23.5
5	√	√	√	√	95.5	85.1	93.9	26.3

* “√” indicates that the current network uses this structure or method; X indicates that the structure or method is not in use on the current network