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

Table 3

Characteristic analysis of loss function.

Loss	Merit	Defect
IoU [16]	Scale invariance, nonnegative, identity, symmetry, triangular inequality	If the anchor frames do not intersect, the distance between the two frames cannot be reflected. Cannot accurately reflect the coincidence of the two boxes
GIoU [17]	A minimum bounding box with zero loss without overlap	When the two boxes intersect, the horizontal and vertical convergence is slow
DIoU [18]	Direct regression of the Euclidean distance between the center points of two frames to accelerate convergence	The aspect ratio is not considered in the regression process
CIoU [19]	Increase detection frame size loss and width loss	The aspect ratio is described as a relative value; Unconsidering difficult sample equilibrium problem