极轻量的航空影像港口船舶目标检测器

Ultra-lightweight detector for port ship targets in aerial images

航空影像下的目标检测由于目标较小,属于该领域中的难点,而通过增加网络的宽度和深度增加其可检能力,会在计算量上付出巨大代价。为此,提供一种极轻量的目标检测器用于航空图像中的目标侦测。算法以港口场景船只目标为例进行仿真,模型大小仅1.64 MB,0.75 T的移动端算力可实现25 fps的速度,同时兼顾准确率。当采用平铺感兴趣区域方式检测时,能够获得足以与超大型网络媲美的准确率。

Target detection in aerial images has always been a difficult point in this field because of the small target.By increasing the width and depth of the network,its detectability is increased,there will be a huge cost in computation.To this end,an extremely lightweight object detector was proposed for object detection in aerial images.The detection of ship targets were taken as an example for simulation.Its model size is only 1.64 MB,and it can achieve a speed of 25 fps on a mobile platform with 0.75 T computing power while taking the accuracy into account.When the method of tiling the region of interest is used for detection,it can obtain accuracy comparable to that of a super-large network.