软件导刊2025，Vol.24Issue(6)：168-174,7.DOI:10.11907/rjdk.241448
基于改进UNet++的洪水区域图像分割算法
Flood Region Image Segmentation Algorithm Based on Improved UNet++
摘要
Abstract
Accurately identifying and delineating areas that may be affected by floods is crucial for planning and implementing flood control measures.A flood area image segmentation algorithm based on improved UNet++is proposed to address the problem of being easily affected by factors such as terrain,climate,lighting,and data imbalance in the process of identifying flood areas.This algorithm is based on the UNet++segmentation network framework,and embeds a hollow space pyramid pooling after the first convolution unit of each decoder layer to obtain new decoding features;At the same time,the features of different decoding paths are fused through dense connections and skip connections,and the Lovasz Hint Loss function is used to obtain the global optimum.The original data is then expanded through data augmentation.The ex-perimental results show that the IOU value of this algorithm on the Flood Area Segmentation dataset reaches 80.53%,which is 7.41,6.21,3.81,3.70,1.92,and 1.82 percentage points higher than the currently popular image segmentation algorithms DeepLabv3,UNet,FCN(Res18),PspNet,FCN(Res50),and UNet++,respectively.The proposed algorithm has high segmentation accuracy and good stability,with excellent overall performance,providing technical support for practical flood control monitoring.关键词
洪水/图像分割/卷积神经网络/UNet++/空洞空间金字塔池化Key words
flood/image segmentation/convolutional neural networks/UNet++/atrous spatial pyramid pooling分类
计算机与自动化引用本文复制引用
姬正杰,魏霖静..基于改进UNet++的洪水区域图像分割算法[J].软件导刊,2025,24(6):168-174,7.基金项目
甘肃省重点研发计划项目(23YFWA003) （23YFWA003）
移动通信国家重点实验室开放课题(2023D15) （2023D15）
