摘要
针对遥感图像分割领域的模糊性和不确定性问题,提出一种全新的遥感图像分割方法。融合了均值和方差模糊以增强模型鲁棒性,采用合页损失函数简化计算,并引入结合遗传算法和基于约束的优化算法的新优化算法,以提升模型性能和准确度。通过多组不同光谱和空间分辨率的遥感数据进行评估,计算了多个有效性指标,并进行详细比较分析。在高尔夫球场和海港地物类别分类实验中,相比于先进的区间二型模糊神经网络等方法,整体准确率平均提高了17.05%和6.5%。文中方法有效解决了遥感图像分割中的模糊性和不确定性问题,为遥感图像分割领域提供一个新的研究思路。
This study is dedicated to tackling the fuzziness and uncertainty inherent in remote sensing image segmentation,presenting a novel approach for the task.The method integrates mean and variance fuzziness to enhance the model's robustness,employs a hinge loss function to simplify computations,and introduces a new optimization algorithm that combines genetic algorithms with constraint-based optimization techniques to boost model performance and accuracy.The evaluation was conducted using multiple sets of remote sensing data with varying spectral and spatial resolutions,calculating several validity indicators for detailed comparative analysis.In experiments involving the classification of golf courses and harbor features,the overall accuracy improved by an average of 17.05%and 6.5%,respectively,compared to advanced methods such as interval type-2 fuzzy neural networks.This approach effectively addresses the fuzziness and uncertainty in remote sensing image segmentation,paving the way for new research avenues in the field.
作者
王春艳
金鹏
桂琪皓
WANG Chunyan;JIN Peng;GUI Qihao(School of Software,Liaoning Technical University,Liaoning 125105,China)
出处
《测绘科学》
CSCD
北大核心
2024年第5期84-98,共15页
Science of Surveying and Mapping
基金
国家自然科学基金-青年科学基金项目(41801368)
辽宁省教育厅基本科研项目(青年项目)(LJKOZ2021154)
关键词
遥感图像分割
模糊性
不确定性
区间二型模糊神经网络
地物类别分类
remote sensing image segmentation
fuzziness
uncertainty
interval type-2 fuzzy neural network
land cover classification
作者简介
王春艳(1981—),女,辽宁阜新人,副教授,博士,主要研究方向为遥感信息识别与提取。E-mail:wangchunyan@lntu.edu.cn;通信作者:金鹏,硕士研究生,E-mail:2396091297@qq.com
