摘要
偏振成像目标检测对于人造目标检测有着重要意义。像元耦合是以四个方向的偏振强度数据作为一个超像元的偏振成像方法。对超像元进行偏振参量解析,会使图像的分辨率变为原始图像的四分之一,不利于小目标的检测。像元耦合图像的偏振参量解析会产生噪声,对小目标的检测造成干扰。本文提出了一个以YOLOv5s为网络基础,添加偏振信息解析模块(Covcat)的目标检测算法。该算法实现了端到端进行像元耦合偏振成像的目标检测,用网络实现偏振解析,利用多卷积信息融合提高特征提取能力,提高目标的平均检测精度(mAP)。使用对空无人机数据集对算法进行验证,实验表明,相比于使用偏振参量解析出的强度图、偏振度图和偏振角图,该算法的平均检测精度分别提升了4个百分点、5个百分点和12个百分点。
Polarization imaging target detection is of great significance for man made target detection.Pixel coupling is a polarization imaging method using polarization intensity data in four directions as a super pixel.The polarization parameters of super pixels reduce the resolution of the image to a quarter of the original image,which is not conducive to the detection of small targets.The polarization parameter analyses of the pixel coupled image generate noise,which can interfere with the detection of small targets.In this paper,a target detection algorithm based on YOLOv5s network with the addition of a polarization information analysis module(Covcat)is proposed.The algorithm achieves end to end pixel coupled polarization imaging target detection,using network for polarization analysis,multi convolution information fusion to improve feature extraction ability,and average detection accuracy(mAP)of targets.The algorithm is verified by using the aerial drone data set,showing that the average detection accuracy of the algorithm is improved by 4 percentage points and 5 percentage points and 12 percentage points compared with the intensity,polarization and polarization angles maps obtained using polarization parameters.
作者
姜黎玮
韩裕生
JIANG Li-wei;HAN Yu-sheng(Department of Information Engineering,PLA Army Artillery Air Defense Force College,Hefei 230031,China;Key Laboratory of Polarized Light Imaging Detection Technology of Anhui Province,Hefei 230031,China)
出处
《激光与红外》
CAS
CSCD
北大核心
2023年第6期970-976,共7页
Laser & Infrared
