遥感舰船目标检测识别方法

针对基于经典图像处理方法的目标检测识别方法虚警率高、分类效果差等问题,提出了一种基于深度学习的光学遥感舰船目标检测识别方法。该方法采用形态学运算+深度学习的方法,基于视觉增强技术快速筛选疑似目标,大幅降低需处理的数据量;采用深度学习网络,大幅降低目标检测虚警率。在2片Xilinx FPGA上完成了设计验证,利用FPGA全并行流水处理的特点,大幅提升处理效率和实时性,相对采用i7-CPU和GPU-GTX1050实现该算法,能效比分别提升260倍和28倍。经16景高分2号卫星遥感图像验证,目标检测识别率高于98%,虚警率低于5%。与现有的目标检测识别方法比,该方法在工程化能力、鲁棒性、实时性、准确率、能效比等方面达到较好平衡,性能优越,优于当前业内方法。

Remote Sensing Ship Target Detection and Recognition Method

Aiming at the problems of high false alarm rate and poor classification effect of the target detection and recognition method based on the classical image processing method,the method of remote sensing ship target detection and recognition based on morphological operations&deep learning is proposed.Through rapid screening of suspected targets based on visual enhancement technology to reduce the amount of data processing greatly,the deep learning network is adopted to reduce false alarm.This proposed method has completed the design verification on two Xilinx FPGA using the features of FPGA full parallel pipelining,and the processing efficiency&real-time performance are greatly improved.Compared with i7-CPU and GPU-GTX1050,the energy efficiency ratio of the algorithm is 260 times and 28 times higher.The verification of 16 high-resolution satellite remote sensing images shows that the recognition rate of target detection is higher than 98%and the false alarm rate is lower than 5%.Compared with the existing methods of target detection and recognition,our method achieves a better balance in engineering capability,robustness,real-time performance,accuracy,and energy efficiency ratio and has superior performance.