多层次特征和粒子群优化的场景分类  

Multi-level feature and particle swarm optimization for scene classification

作　　者：张立亭[1] 喻欣 罗亦泳[1] 杨静雯 ZHANG Li-ting;YU Xin;LUO Yi-yong;YANG Jing-wen(School of Geomatics,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学测绘工程学院,江西南昌330013

出　　处：《计算机工程与设计》2023年第9期2747-2753,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(41861058)。

摘　　要：针对遥感图像的场景分类精度问题,提出多层次特征和粒子群算法优化分类器的场景分类算法。利用聚集局部描述符编码算法对尺度不变特征变换算法提取的局部特征编码,获得中层特征,通过卷积神经网络提取高层特征,将提取的特征作为支持向量机的输入数据,引入粒子群算法优化该分类器的参数,进行场景分类。在RSC11和WHU-RS19两个公开的遥感图像数据集上进行实验,分类精度分别达到95.28和97.20。将WHU-RS19数据集的结果与其它方法比较,精度有明显提高。实验结果表明,在分类时对分类器参数进行优化,分类效果更佳。Aiming at the problem of scene classification accuracy of remote sensing image,a scene classification algorithm optimized by multi-layer feature and particle swarm optimization was proposed.The local descriptor gathered coding algorithm was used to extract the local characteristics coding of scale invariant feature transform algorithm,middle-level characteristics were obtained.Through the convolutional neural network,high-level features were extracted as the input data of support vector machine,the particle swarm algorithm was introduced to optimize the parameters of the classifier,and the scene classification was realized.Experiments were carried out on two public remote sensing image datasets,RSC11 and WHU-RS19.Results show that the classification accuracies reach 95.28 and 97.20,respectively.The results of WHU-RS19 dataset are compared with other methods,and the accuracy is improved obviously.Experimental results show that the classification effect is better when the parameters of the classifier are optimized.

关 键 词：遥感图像 场景分类 尺度不变特征变换 聚集局部描述符编码算法 卷积神经网络 支持向量机 粒子群算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]