摘要
针对卫星影像上车辆的漏检问题,该文对深度学习YOLOv3模型进行了网络改进,并用于高分二号卫星影像车辆检测。该方法通过减少原特征提取网络darknet-53的层数来降低网络复杂度,并在原YOLOv3模型3个尺度的基础上进行了尺度扩充以提高对小目标的检测能力。实验结果表明,改进后的YOLOv3模型较好地克服了多数深度学习算法不擅长检测小目标的短板,不仅检测结果比原算法更为精确,而且训练和检测速度也更快,具有一定的优势。
Aiming at the problem of missing detection of vehicles on satellite images,this paper improved the deep learning model YOLOv3 and used it for vehicle detection in Gaofen-2 satellite images.This method reduced the model complexity by reducing the number of layers of the original feature extraction model darknet-53,and expanded the scale based on the original YOLOv3 model to improve the detection ability of small targets.Experimental results showed that the improved YOLOv3 model better overcomes the shortcomings of most deep learning algorithms that are not good at detecting small targets.Not only is the detection result more accurate than the original algorithm,but the training and detection speed is also faster,which has certain advantages.
作者
彭新月
张吴明
钟若飞
PENG Xinyue;ZHANG Wuming;ZHONG Ruofei(State Key Laboratory of Remote Sensing Science»Beijing Normal University,Beijing 100875,China;School of Geospatial Engineering and Science,Sun Yat-sen University,Zhuhai,Guangdong 519082,China;Southern Marine Science and Engineering Guangdong Laboratory(Zhuhai).Zhuhai,Guangdong 519080,China;Key Laboratory of 3-Dimensional Information Acquisition and Application,Ministry of Education,Capital Normal University,Beijing 100048,China)
出处
《测绘科学》
CSCD
北大核心
2021年第12期147-154,共8页
Science of Surveying and Mapping
基金
国家自然科学基金面上项目(41971380,41671414,42071444)
广西自然科学基金—创新研究团队项目(2019GXNSFGA245001)
遥感科学国家重点实验室开放基金项目(OFSLRSS201920)。
