基于深度学习的航空影像非正规垃圾堆放点监测技术研究与实践

Research and practice on the monitoring technology of informal rubbish heaps in aerial images based on deep learning

城市化进程的加快导致垃圾随处堆放的问题日益突出,给城市的环境及居民的生活质量造成了严重的影响。利用遥感手段快速监测非正规垃圾堆放点具有及时性和高效性,因此具有十分重要的意义。本文结合无人机高分辨率航空影像及非正规垃圾堆分布特征,提出了按地域特征勾画样本数据集提取样本数据特征,采用U-Net和Swin Transformer融合模型,以及针对性改进训练流程开展非正规垃圾堆放点信息分类研究。试验以绍兴市越城区、柯桥区和上虞区作为研究区域,利用飞马航测无人机获取航空影像数据,对比分析了本文提出的方法和基于深度学习的典型地物要素提取方法在非正规垃圾堆放点监测上的应用,试验结果表明本文提出的方法准确率提高了1.72倍。

The fast-growing urbanization has led to the increasingly problem of rubbish heaps everywhere,which cause a serious impact on the urban environment and the quality life of residents.It has significant influence to monitor informal rubbish heaps by using remote sensing in a timely and efficient manner.With using the high resolution aerial images of UAVs and the distribution features of informal rubbish heaps,this paper proposes a region layer-based method to sample labelling rubbish heaps,and uses the U-Net and Swin Transformer fusion model and the improvement of the training process to carry out the research on the classification of informal rubbish heaps.The experiment takes Shangyu district,Keqiao District and Yuecheng District of Shaoxing city as the research area,uses the Feima aerial survey UAV to obtain aerial image.By comparing the proposed method in this study and the typical ground objects extraction based on Deep Learning in the monitoring of informal rubbish heaps.The results show that the accuracy of the proposed method in this study is 1.72 times higher than typical ground objects extraction.