复杂条件下小目标检测算法研究
|
摘要
非海空背景小目标检测是图像处理最具挑战的任务之一。为了解决复杂条件下的小目标检测准确率不足的情况,本文提出首先运用超分辨率模型对拍摄模糊图像进行重建,将重建后的清晰图像进行小目标检测。另外,对原始FPN模型进行改进,利用浅层网络丰富的位置信息,仅采用三层特征提取网络,即可完成小目标全图搜索检测。实验表明,本文方法在清晰图像直接进行重建准确率达到81.82%,map值为0.895 1,重建后的再进行小目标检测与清晰图像直接检测仅有一个未检测出。
Small target detection is one of the most challenging tasks in image processing. In order to solve the problem of insufficient accuracy of small target detection under complex conditions,this paper proposes to reconstruct the captured blurred image by using the super-resolution model,and to detect the reconstructed clear image for small target. In addition,the original FPN model is improved,and rich location information of the shallowtarget network is used,and only the three-layer feature extraction network is used to complete the small target full-image search detection. Experiments showthat the accuracy of this method is 81.82% and the map value is 0.895 1. The reconstructed small target detection and clear image direct detection have only one undetected difference.
Small target detection is one of the most challenging tasks in image processing. In order to solve the problem of insufficient accuracy of small target detection under complex conditions,this paper proposes to reconstruct the captured blurred image by using the super-resolution model,and to detect the reconstructed clear image for small target. In addition,the original FPN model is improved,and rich location information of the shallowtarget network is used,and only the three-layer feature extraction network is used to complete the small target full-image search detection. Experiments showthat the accuracy of this method is 81.82% and the map value is 0.895 1. The reconstructed small target detection and clear image direct detection have only one undetected difference.
引文
[1] GIRSHICK R,DONAHUE J,DARRELL T,et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision&Pattern Recognition. Columbus,OH,USA:IEEE,2014:580-587.
[2] HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[M]//FLEET D,et al. ECCV 2014,Part III,LNCS8691.Switzerland:Springer International Publishing,2014:346-361.
[3] GIRSHICK R. Fast R-CNN[J]. ar Xiv preprint ar Xiv:1504.08083,2015.
[4] REN Shaoqing,HE Kaiming,GIRSHICK R,et al. Faster RCNN:Tow ards real-time object detection w ith region proposal netw orks[C]//IEEE Transactions on Pattern Analysis&M achine Intelligence,2017,39(6):1137-1149.
[5]LIU Wei,ANGUELOV D,ERHAN D,et al. SSD:Single shot multibox detector[J]. ar Xiv preprint ar Xiv:1512.02325,2015.
[6]REDMON J,DIVVALA S,GIRSHICK R,et al. You only look once:Unified,real-time object detection[J]. ar Xiv preprint ar Xiv:1506.02640,2015.
[7] REDMON J,FARHADI A. YOLO9000:Better,faster,stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Honolulu,Haw aii,USA:IEEE,2017:6517-6525.
[8]朱国强,孟祥勇,钱惟贤.基于曲率的近地面红外小目标检测算法[J].光子学报,2018,47(10):1010001(1-12).
[9]胡洪涛,敬忠良,胡士强.基于辅助粒子滤波的红外小目标检测前跟踪算法[J].控制与决策,2005,20(11):1208-1211.
[10]王军,姜志,孙慧婷,等.基于噪声方差估计的红外弱小目标检测与跟踪方法[J].光电子·激光,2018,29(3):305-313.
[11]于晓涵,陈小龙,陈宝欣,等.快速高分辨稀疏FRFT雷达机动目标检测方法[J].光电工程,2018,45(6):170702(1-7).
[12]李东,赵婷,宋伟,等.一种低信噪比下稳健的ISAR平动补偿方法[J].电子学报,2018,46(9):2049-2056.
[13]LIN T Y,DOLLAR P,GIRSHICK R,et al. Feature pyramid netw orks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Honolulu,Haw aii,USA:IEEE Computer Society,2017:936-944.
[14]TAO Xin,GAO Hongyun,WANG Yi,et al. Scale-recurrent netw ork for deep image deblurring[J]. ar Xiv preprint ar Xiv:1802.01770,2018.
[15]DONG Chao,LOY C C,HE Kaiming,et al. Learning a deep convolutional netw ork for image super-resolution[M]//FLEET D,et al. ECCV 2014,Part IV,LNCS 8692.Sw itzerland:Springer International Publishing,2014:184-199.
[16]DONG Chao,LOY C C,TANG Xiaoou. Accelerating the super-resolution convolutional neural netw ork[J]. ar Xiv preprint ar Xiv:1608.00369,2016.
[17]SHI Wenzhi,CABALLERO J,HUSZR F,et al. Real-time single image and video super-resolution using an efficient subpixel convolutional neural netw ork[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Las Vegas,NV,USA:IEEE,2016:1874-1883.
[2] HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[M]//FLEET D,et al. ECCV 2014,Part III,LNCS8691.Switzerland:Springer International Publishing,2014:346-361.
[3] GIRSHICK R. Fast R-CNN[J]. ar Xiv preprint ar Xiv:1504.08083,2015.
[4] REN Shaoqing,HE Kaiming,GIRSHICK R,et al. Faster RCNN:Tow ards real-time object detection w ith region proposal netw orks[C]//IEEE Transactions on Pattern Analysis&M achine Intelligence,2017,39(6):1137-1149.
[5]LIU Wei,ANGUELOV D,ERHAN D,et al. SSD:Single shot multibox detector[J]. ar Xiv preprint ar Xiv:1512.02325,2015.
[6]REDMON J,DIVVALA S,GIRSHICK R,et al. You only look once:Unified,real-time object detection[J]. ar Xiv preprint ar Xiv:1506.02640,2015.
[7] REDMON J,FARHADI A. YOLO9000:Better,faster,stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Honolulu,Haw aii,USA:IEEE,2017:6517-6525.
[8]朱国强,孟祥勇,钱惟贤.基于曲率的近地面红外小目标检测算法[J].光子学报,2018,47(10):1010001(1-12).
[9]胡洪涛,敬忠良,胡士强.基于辅助粒子滤波的红外小目标检测前跟踪算法[J].控制与决策,2005,20(11):1208-1211.
[10]王军,姜志,孙慧婷,等.基于噪声方差估计的红外弱小目标检测与跟踪方法[J].光电子·激光,2018,29(3):305-313.
[11]于晓涵,陈小龙,陈宝欣,等.快速高分辨稀疏FRFT雷达机动目标检测方法[J].光电工程,2018,45(6):170702(1-7).
[12]李东,赵婷,宋伟,等.一种低信噪比下稳健的ISAR平动补偿方法[J].电子学报,2018,46(9):2049-2056.
[13]LIN T Y,DOLLAR P,GIRSHICK R,et al. Feature pyramid netw orks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Honolulu,Haw aii,USA:IEEE Computer Society,2017:936-944.
[14]TAO Xin,GAO Hongyun,WANG Yi,et al. Scale-recurrent netw ork for deep image deblurring[J]. ar Xiv preprint ar Xiv:1802.01770,2018.
[15]DONG Chao,LOY C C,HE Kaiming,et al. Learning a deep convolutional netw ork for image super-resolution[M]//FLEET D,et al. ECCV 2014,Part IV,LNCS 8692.Sw itzerland:Springer International Publishing,2014:184-199.
[16]DONG Chao,LOY C C,TANG Xiaoou. Accelerating the super-resolution convolutional neural netw ork[J]. ar Xiv preprint ar Xiv:1608.00369,2016.
[17]SHI Wenzhi,CABALLERO J,HUSZR F,et al. Real-time single image and video super-resolution using an efficient subpixel convolutional neural netw ork[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Las Vegas,NV,USA:IEEE,2016:1874-1883.