液晶与显示, 2019, 34 (4): 439, 网络出版: 2019-06-12
Faster-RCNN和Level-Set结合的高分遥感影像建筑物提取
High-resolution remote sensing image building extraction combined with Faster-RCNN and Level-Set
摘要
目前Level-Set图像分割方法存在初始轮廓的确定受人为因素影响较大的问题, 对目标被遮盖和目标与背景灰度值相近无法达到理想的分割效果。针对此问题, 本文提出了利用Faster-RCNN网络模型确定目标初始轮廓和区域信息的先验水平集图像分割方法, 搭建Caffe深度学习框架训练Faster-RCNN网络模型; 通过有监督学习的方式在IAILD数据集上训练模型, 检测出目标建筑物并初步提取建筑物的轮廓, 并将其与形状先验的Level-Set算法结合。对比实验结果表明, 本文方法解决了Level-Set算法中图像分割结果初始轮廓受人为标记框选的影响较大的问题, 能够更好地完成被遮挡建筑物的分割, 对于目标建筑和背景灰度值相近也能达到更好的分割效果。
Abstract
At present, the Level-Set image segmentation method has the problem that the initial contour is greatly affected by human factors. The target segmentation and the target and background gray values are close to each other, and the ideal segmentation effect cannot be achieved. Aiming at this problem, this paper proposes a priori level set image segmentation method based on the Faster-RCNN network model to determine the initial contour and region information of the target, and builds the Caffe deep learning framework to train the Faster-RCNN network model. The IAILD data is obtained through supervised learning. The training model is assembled, the target building is detected and the outline of the building is initially extracted and combined with the shape-priority Level-Set algorithm. The experimental results show that the proposed method solves the problem that the initial contour of the image segmentation result is greatly influenced by the artificial marker frame selection in the Level-Set algorithm. It can better complete the segmentation of the occluded building, and achieve better segmentation effect for the target building and the background gray value.
左俊皓, 赵聪, 朱晓龙, 任洪娥. Faster-RCNN和Level-Set结合的高分遥感影像建筑物提取[J]. 液晶与显示, 2019, 34(4): 439. ZUO Jun-hao, ZHAO Cong, ZHU Xiao-long, REN Hong-e. High-resolution remote sensing image building extraction combined with Faster-RCNN and Level-Set[J]. Chinese Journal of Liquid Crystals and Displays, 2019, 34(4): 439.