首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 223 全文下载次数: 138
引用本文:

DOI:

10.11834/jrs.20210163

收稿日期:

2020-05-17

修改日期:

2020-07-22

PDF Free   EndNote   BibTeX
结合多光谱影像降维与深度学习的城市单木树冠检测研究
奚祥书, 夏凯, 杨垠晖, 杜晓晨, 冯海林
浙江农林大学
摘要:

多光谱数据的降维处理对基于深度学习的单木树冠检测研究有重要意义,如何使用合适的降维方法以提高 单木检测的精度却少有研究讨论。本文使用无人机搭载多光谱相机进行航拍作业,采集研究区内银杏树种多光谱 影像。将原始多光谱影像通过特征波段选择、特征提取、波段组合的方法生成 5 种不同的数据集用于训练 3 种经 典的深度学习网络 FPN-Faster-R-CNN,YOLOv3,Faster R-CNN。其中由波段组合方法得到的近红外、红色、绿色 波段组合在不同类型的目标检测网络中都有最好的检测结果,其中 FPN-Faster-R-CNN 网络对银杏树冠的检测精度 最高为 88.4%,由 OIF 指标得到的蓝色、红色、近红外波段组合信息量最高,但在所有网络中的平均检测精度最 低,仅为 79.3%。实验结果表明:在不同波段降维方法中,若降维后的影像中目标物体的色彩与背景差异较明显, 且轮廓清晰,则深度学习网络对树冠的检测可获得较好的结果。而影像自身的信息量则对深度学习网络的树冠检 测能力的提升作用有限。本研究中针对多光谱影像的降维方法分析,为基于深度学习的单木树冠检测研究提供了 重要的实验参考。

Urban individual tree crown detection research using multispectral images dimensionality reduction with deep learning
Abstract:

The dimensionality reduction processing of multi-spectral data is of great significance to the deep learning-based single-tree crown detection research. However, how to use the appropriate dimensionality reduction method to improve the accuracy of single-tree detection is rarely discussed. In this work, a multi-spectral camera equipped with uav was used for aerial photography to collect multi-spectral images of ginkgo tree species in the research area. The original multi-spectral images were used to generate 5 different data sets through feature band selection , feature extraction ,band combination method for training 3 classical deep learning networks: FPN-Faster-R-CNN, YOLOv3, Faster R-CNN. Which by the characteristic of bands selection method, red and green ,near infrared bands combination in different types of target detection in the network has the best results, the FPN-Faster-R-CNN network detection accuracy up to 88.4% of ginkgo tree. Blue, red, near infrared band combination which are obtained by OIF index the highest amount of information, but in all the lowest average accuracy of network, is only 79.3%. The experimental results show that: in the different dimensionality reduction methods , if the color and background of the target object in the image after dimensionality reduction are obviously different, and the contour is clear, the deep learning network can obtain better results in the detection of tree crown. However, the information content of the image itself has limited effect on the ability of deep learning network to detect tree crown. In this study, the dimensionality reduction method of multi-spectral images is analyzed, which provides an important experimental reference for the deep learning based single tree crown detection.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮