A Surface Target Recognition Algorithm Based on Coordinate Attention and Double-Layer Cascade

Table 4

Accuracy for each category on VOC2012.

Model	Backbone	GPU	Aero	Bike	Boat	Bus	Car	mAP
Fast R-CNN	VGG-16	Yes	82.3	78.4	52.3	77.8	71.6	68.4
Faster R-CNN	ResNet-101	Yes	84.9	79.8	53.9	77.5	75.9	70.4
YOLO	DarkNet-53	Yes	77.0	67.2	38.3	68.9	55.9	57.9
SSD300	VGG-16	Yes	85.6	80.1	57.6	79.4	76.1	76.8
SSD513	ResNet-101	Yes	87.4	82.3	59.0	81.7	81.5	74.9
YOLOv3	DarkNet-53	Yes	81.7	85.7	65.5	87.2	87.5	74.5
Our model	CSP DarkNet-53	Yes	85.3	81.9	76.2	88.9	83.6	78.2