结合水体指数与卷积神经网络的遥感水体提取

发布时间：2018-01-15 16:35

本文关键词：结合水体指数与卷积神经网络的遥感水体提取　出处：《遥感信息》2017年05期 　论文类型：期刊论文

【摘要】：常用多光谱遥感水体提取少有兼顾光谱与空间信息,致使水体提取的可靠性和准确性难以保证。在利用遥感水体光谱特性的同时,融入深度学习算法,提出归一化差分水体指数(normalized difference water index,NDWI)与深度学习联合的遥感水体提取方法。该方法首先选取典型水体样本进行训练,构建深度学习卷积神经网络(convolutional neural networks,CNN)水体识别模型。其次,计算多光谱影像NDWI指数并分割成图斑,以图斑包络矩形构建初始的水体目标子区。最后,构建NDWI指数与CNN水体识别概率的联合估计模型,并以迭代运算实现最优化遥感水体提取。实验验证了该方法的高可靠性与准确性。相比常用方法,水体识别准确率高达94.19%,而错分率仅为5.04%,显著提高了水体提取精度。
[Abstract]:It is difficult to guarantee the reliability and accuracy of the multi-spectral remote sensing water body extraction, which makes it difficult to ensure the reliability and accuracy of the multi-spectral remote sensing water body extraction. At the same time, the depth learning algorithm is incorporated into the remote sensing water body spectral characteristics. A normalized difference water index is proposed. NDWI) and depth learning method for remote sensing water body extraction. Firstly, the typical water samples are selected for training. Construct the model of convolutional neural networks and CNNs. Secondly. The NDWI exponent of multispectral image is calculated and divided into speckles, and the initial target area of water body is constructed by using the image patch envelope rectangle. Finally, the joint estimation model of NDWI exponent and CNN water body recognition probability is constructed. Iterative operation is used to optimize remote sensing water extraction. The experimental results show that this method has high reliability and accuracy. Compared with common methods, the accuracy of water body recognition is as high as 94.19%. However, the error rate was only 5.04, which significantly improved the precision of water extraction.
【作者单位】：东华理工大学测绘工程学院;流域生态与地理环境监测国家测绘地理信息局重点实验室;武汉大学资源与环境科学学院;东华理工大学水资源与环境工程学院;
【基金】：国家自然科学基金(41401526) 江西省自然科学基金(20171BAB213025) 流域生态与地理环境监测国家测绘地理信息局重点实验室资助课题(WE2015003) 江西省教育厅科技项目江西省高等学校科技落地计划项目(KJLD14049)
【分类号】：TP183;TP751
【正文快照】： 0引言地表水体的形成与扩张、萎缩与消失是区域气候变化和生态环境演化的重要影响因素[1]。精准地提取水体信息,为区域生态环境时空演化规律研究提供必要的数据支撑,对于水资源调查、水利规划、流域治理、洪水监测与灾后评估具有重要意义。迄今为止,国内外许多研究人员对遥感

【相似文献】