一种融合RF-SVM的无人机遥感图像信息提取与分类技术  

Information Extraction and Classification Technology of UAV Remote Sensing Image Based on RF-SVM

作　　者：黄净晴[1] 王浩[2] HUANG Jing-qing;WANG Hao(Fuzhou University of Electronics and Communications Engineering,Fuzhou 350001,China;Fuzhou University,Fuzhou 350001,China)

机构地区：[1]福建信息职业技术学院,福建福州350001 [2]福州大学,福建福州350001

出　　处：《长春师范大学学报》2023年第4期66-71,共6页Journal of Changchun Normal University

基　　金：水电科学研究院流域水循环模拟与调控2018年度国家重点实验室开放基金项目“推移质运动与紊流结构作用关系试验研究”(IWHR-SKL-KF201815)。

摘　　要：针对传统无人机遥感图像信息提取与分类算法准确率低、稳定性差、无法有效应对大规模复杂遥感图像数据集等问题,提出一种基于RF-SVM的遥感图像处理算法。RF-SVM算法将RF数据集分类性能较强的优势与经典SVM算法数据降维能力相融合,引入随机变量和示性函数扩大样本集的边界,提升对复杂大规模数据集的处理能力,有效控制泛化误差。在对无人机遥感图像的预处理过程中,借助Brovey变换完成对光谱和高分辨率遥感图像的像素级融合,引入核函数并根据获取到的遥感图像特征和后验概率值,实现对遥感图像内部标的物的准确分类。实验结果显示,在RF-SVM算法下,无人机遥感图像信息提取准确率分类平均准确率达到99.81%,且在RF-SVM算法下的样本点感受性曲线稳定性更好。In view of the low accuracy and poor stability of traditional UAV remote sensing image information extraction and classification algorithms,which can not effectively deal with large-scale and complex remote sensing image data sets,a remote sensing image processing algorithm based on RF-SVM is proposed.RF-SVM algorithm combines the advantages of strong classification performance of RF data sets with the data dimension reduction ability of classical SVM algorithm,introduces random variables and indicative functions to expand the boundary of sample sets,improves the processing ability of complex large-scale data sets,and effectively controls the generalization error.In the preprocessing process of UAV remote sensing images,the pixel level fusion of spectral and high-resolution remote sensing images is completed by means of Brovey transform.The kernel function is introduced and the accurate classification of the internal objects in the remote sensing images is realized according to the acquired remote sensing image features and posterior probability values.The experimental results show that under the RF-SVM algorithm,the average accuracy of UAV remote sensing image information extraction and classification is 99.81%,and the sensitivity curve of sample points under the RF-SVM algorithm is more stable.

关 键 词：RF-SVM 无人机 遥感图像 信息提取 后验概率 Brovey变换 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]