«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

期数:

2025年10期

页码:

181-190

栏目:

新型计算应用系统

出版日期:

2025-10-10

文章信息/Info

Title:

Detection of Degraded Alpine Meadow Bald Patches Based on Optimized SVM with Improved GWO

文章编号:

1673-629X(2025)10-0181-10

作者:

罗 维1; 2; 王 璐1; 2*; 姚 斌1; 2; 郑 敏3; 李希来3

1. 青海大学 计算机技术与应用学院,青海西宁 810016; 2. 青海大学 青海省智能计算与应用实验室,青海西宁 810016; 3. 青海大学 农牧学院,青海西宁 810016

Author(s):

LUO Wei1; 2; WANG Lu1; 2*; YAO Bin1; 2; ZHENG Min3; LI Xi-lai3

1. School of Computer Technology and Application,Qinghai University,Xining 810016,China; 2. Qinghai Provincial Laboratory for Intelligent Computing and Application,Qinghai University,Xining 810016,China; 3. School of Agriculture and Animal Husbandry,Qinghai University,Xining 810016,China

关键词:

灰狼优化算法; 支持向量机; 高光谱成像技术; 特征融合; 高寒草甸秃斑块

Keywords:

gray wolf optimization algorithm; support vector machine; hyperspectral imaging technique; feature fusion; alpine meadow bald patches

分类号:

TP391

DOI:

10.20165/j.cnki.ISSN1673-629X.2025.0149

摘要:

准确识别青藏高原退化高寒草甸秃斑块,有助于评估高寒草甸退化状况,从而为制定科学合理的恢复与管理策略提供数据支持。为了实现对退化高寒草甸秃斑块的快速、无损检测,基于退化高寒草甸秃斑块的高光谱图像,构建了一种改进灰狼优化算法优化的支持向量机模型。首先,收集了不同秃斑块的高光谱图像,并从中提取了高光谱反射率和植被指数。其次,利用CARS和ReliefF算法筛选出关键波长和显著植被指数。基于融合特征数据集(即关键波长加上显著植被指数)的支持向量机(SVM)模型展示出优异的分类性能。再次,基于融合特征数据集,建立了灰狼优化算法优化的支持向量机(GWO-SVM)。最后,鉴于灰狼优化算法(GWO)存在的局限,通过引入拉丁超立方采样和多样性保持策略构建了改进的灰狼优化算法(LDGWO),旨在提升算法的搜索效率。结果表明,LDGWO-SVM在秃斑块检测中达到了96.97%的分类准确率,建模时间为8.41秒,优于GWO-SVM模型。该模型在识别退化高寒草甸秃斑块方面展现了良好的效果,为退化高寒草甸秃斑块的无损检测提供了新的思路和方法。

Abstract:

Accurate identification of degraded alpine meadow bald patches on the Tibetan Plateau is crucial for assessing their degradation status and informing the development of effective restoration and management strategies. To enable rapid and non-invasive detection,a Support Vector Machine (SVM) model optimized with an enhanced Gray Wolf Optimization (GWO) algorithm was developed using hyperspectral images of degraded alpine meadow bald patches. Firstly,hyperspectral images of different bald patches were collected,from which hyperspectral reflectance and vegetation indices were extracted. Secondly,key wavelengths and significant vegetation indices were filtered using CARS and ReliefF algorithms. The SVM model,based on the fused feature dataset (key wavelengths and significant vegetation indices),demonstrated excellent classification performance. Thirdly,a support vector machine optimized by the Gray Wolf Optimization (GWO-SVM) was built based on this fused feature dataset. Finally,to address the limitations of the standard GWO algorithm,an improved version (LDGWO) was constructed by incorporating Latin hypercube sampling and diversity maintenance strategies,aiming to enhance the search efficiency. The results indicate that the LDGWO-SVM model achieves 96.97% classification ac-curacy in detecting bald patches,with a modeling time of 8.41 seconds,outperforming the GWO-SVM model. The proposed model demonstrated good results in recognizing bald patches in degraded alpine meadows, providing new ideas and methods for non-destructive detection of such areas.

常用功能

更新日期/Last Update: 2025-10-10