A Robust and Lightweight Detector for Ship Target with Complex Background in SAR Image

Table 10

Performance comparison of SSDD with different algorithms.

Method	Backbone	Precision (%)	Recall (%)	AP (%)	FLOPs (GFLOs)	params	Runtimes(ms)
Libra R-CNN [23]	ResNet 101-FPN	88.6	88.6	89.9	83.0	60.4	30.2
Cascade R-CNN [24]	ResNet 101-FPN	94.3	89.9	89.5	110.4	87.9	38.8
Faster R-CNN	ResNet 101-FPN	90.9	87.6	88.3	82.7	60.1	30.2
CR2A-Net [25]	ResNet 101-FPN	94.0	87.8	89.8	112.0	88.6	67.2
DAPN [26]	ResNet 101-FPN	87.6	91.4	90.1	117.2	63.8	34.5
RetinaNet [27]	ResNet-101-FPN	81.6	92.3	89.6	71.8	55.1	30.2
SSD	SSD-VGG	92.9	88.0	94.0	87.7	24.4	30.2
YOLOv3	Darknet-53	90.7	94.7	95.0	49.6	61.5	10.4
YOLOv4	CSPDarknet-53	93.6	94.0	96.1	45.3	64.3	12.9
YOLOv5	CSPDarknet-53	93.9	92.8	97.2	16.4	7.1	8.8
[22]	Darknet-53	95.1	94.5	94.8	50.6	65.8	16.4
FCOS [28]	ResNet 101-FPN	94.4	85.6	88.7	69.8	50.8	25.9
CenterNet [29]	DAL-34	93.3	94.5	93.5	25.9	20.2	21.5
CenterNet++ [30]	DAL-34	92.6	94.5	92.7	26.6	20.3	21.5
Our	CSPDarknet-53	98.1	97.1	99.2	11.3	1.9	5.1
Our(0.8 prune)	CSPDarknet-53	94.7	92.0	97.1	0.5	0.1	2.9