摘要
由于数据的缺乏,针对高分辨率遥感影像中的圆形喷灌农田目标检测的研究较少,为此提出基于改进YOLOv3的圆形农田检测方法。基于高分辨率遥感影像建立圆形农田数据集;构建深度学习目标检测模型,选定单阶段检测模型YOLOv3并作如下改进:用Resnet替换骨干网中原本的Darknet53特征提取网络;将检测头中的单向特征金字塔改进为双向特征金字塔;应用Dropblock正则化,综合考虑改进方案,最终搭建RD_YOLOv3检测模型。实验结果表明,在公开遥感数据集HRRSD上,RD_YOLOv3的mAP为86.72%,相对于原版YOLOv3可提高1.27%。在我国西北地区圆形灌溉农田数据集上,RD_YOLOv3的mAP可达92.38%,具有较好的应用效果。该研究可以为遥感图像目标检测和生态治理乡村振兴提供一定的参考价值。
Due to the lack of data,there are fewer studies on detection of the round sprinkling irrigation farmlands in the high⁃resolution remote sensing images.Therefore,a round farmland detection method based on improved YOLOv3 is proposed.In this study,a round farmland dataset is established based on high⁃resolution remote sensing images.An object detection model based on deep learning is established.The single⁃stage detection model YOLOv3 is selected and improved as follows.The original Darknet53 feature extraction network in the backbone network is replaced with Resnet,the unidirectional feature pyramid in the detection head is improved to get a bi⁃directional feature pyramid,and the improved scheme is taken into account comprehensively to build the RD_YOLOv3 detection model by means of Dropblock regularization.The experimental results show that the mAP of RD_YOLOv3 on the public remote sensing data set HRRSD is 86.72%,which is increased by 1.27%in comparison with that of the original YOLOv3.In the round irrigation farmland data set of the northwest China,the mAP of RD_YOLOv3 can reach 92.38%,which has a good application effect.Therefore,this study can provide a certain reference value for remote sensing image object detection and ecological governance in the rural revitalization.
作者
程坤
张斌
郭新
CHENG Kun;ZHANG Bin;GUO Xin(School of Geography and Information Engineering,China University of Geosciences,Wuhan 430074,China;School of Computer Science&Engineering,Wuhan Institute of Technology,Wuhan 430205,China)
出处
《现代电子技术》
2022年第11期24-28,共5页
Modern Electronics Technique
基金
国家自然科学基金项目:基于高层特征学习的高分辨率遥感影像农村居民点土地利用分类(41601480)。
