基于深度神经网络的河流遥感图像分割方法研究  

Research on river remote sensing image segmentation method based on deep neural network

作　　者：李宗斌[1] LI Zongbin(Chongqing University of Arts and Sciences,Chongqing 402160,China)

机构地区：[1]重庆文理学院,重庆402160

出　　处：《人民长江》2024年第7期73-78,97,共7页Yangtze River

基　　金：国家自然科学基金区域创新发展联合基金项目(U22A20102)。

摘　　要：为解决河流遥感图像分割效果较差且交并比较低的问题,提出了基于深度神经网络的河流遥感图像分割方法。通过对高空间分辨率的河流遥感图像数据集的分析,预处理河流遥感图像,解决数据集中存在的弱标签问题;采用卷积编码-解码网络构建深度神经网络的特征提取模型,并运用KNN算法实现河流遥感图像的高精度分割;最后以重庆市嘉陵江2022年河流遥感图像为例进行验证。实验结果表明:所提方法能够保留分割后的图像细节特征,且图像分割交并比较高,为0.94。所提方法能够对河流遥感图像进行高精度分割,可为水资源管理和环境保护等方面提供技术支持。To solve the problem of poor segmentation effects and low intersection and union ratio of river remote sensing images,a river remote sensing image segmentation method based on the deep neural network was proposed.By analyzing the river remote sensing image data set with high spatial resolution,the river remote sensing image was preprocessed to solve the weak label problem in the data set;the convolutional coding-decoding network was used to construct a feature extraction model of deep neural network,and the KNN algorithm was used to realize the high-precision segmentation of river remote sensing images;finally,the remote sensing images of the Jialing River in Chongqing City in 2022 was taken as an example for verification.The experimental results showed that the proposed method can preserve the detailed features of the segmented image,and the intersection and union ratio of image segmentation was high,which was 0.94.The proposed method can achieve high-precision segmentation of river remote sensing images,providing technical support for water resource management and environmental protection.

关 键 词：河流遥感图像 图像分割 特征提取 残差连接 深度神经网络 嘉陵江 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]