A SAR image ship detection method combined with anchor-free and attention module
Xin Hu, Lijun Ma, Diyuan Xiang, Hao Qin · International Symposium on Robotics, Artificial Intelligence, and Information Engineering (RAIIE 2022) · 2022
The detection of SAR images of ship recently become one of the important research direction, but the SAR image imaging feature is not obvious, the face of a complex scenario testing result is not ideal, can produce serious leakage inspection by mistake, and so on and so forth, in addition, the deployment in the actual situation, the detection accuracy and speed cannot be both at the same time, lead to much difficulty in the practical application. This paper designed a anchor-free detection algorithm TA-FCOS on the basis of FCOS, which can effectively reduce the number of parameters. Meanwhile, we proposed an attention module TA, which was embedded into the backbone network to effectively improve the detection accuracy. Finally, we conducted experiments on the HRSID data set. AP50 and mAP reach 85.9% and 57.8%, respectively. Experimental results confirm the effectiveness of our algorithm.