基于多特征融合与Fisher准则的遥感场景图像分类  

Remote sensing scene image classification based on multi-feature fusion and the Fisher criterion

作　　者：高翔[1] 侯宇超 程蓉 续婷[1] 王李祺 白艳萍[1] GAO Xiang;HOU Yu-chao;CHENG Rong;XU Ting;WANG Li-qi;BAI Yan-ping(School of Mathematics,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学,数学学院,太原030051

出　　处：《兰州大学学报（自然科学版）》2023年第1期90-97,共8页Journal of Lanzhou University(Natural Sciences)

基　　金：国家自然科学基金项目(61774137,51875535,61927807);山西省重点研发计划项目(201903D121156);山西省回国留学人员科研项目(2020-104,2021-108)。

摘　　要：提出一种基于多特征融合的Fisher准则分类方法,将提取到的卷积神经网络特征、图像纹理的局部二值模式、方向梯度直方图特征及颜色特征、颜色矩进行有效融合,使其在高维空间上线性可分,利用线性分类器Fisher对特征模型的参数进行微调获得分类结果.将该模型应用于数据集UCM进行测试,与其他分类方法相比,准确率均有所提升;与深度卷积网络GoogLeNet相比准确率提升1.5%.为保证该模型的泛用性,于AID数据集上进行进一步实验,结果验证了该模型的有效性.A Fisher criterion classification method based on multi-feature fusion was proposed,in which the extracted convolutional neural network features,local binary pattern and histogram of oriented gradient features of image textures and color feature color moments were effectively fused to make them linearly separable in a high-dimensional space,and the parameters of the feature model were fine-tuned via the linear classifier Fisher to obtain the classification results.The model was applied to the public dataset UCM for testing,which improved the accuracy of various classification methods proposed in literature in recent years,and also improved by 1.5%as against the deep convolutional network GoogLeNet.In order to ensure the versatility of the model,further experiments were carried out on the AID dataset,and the results also verified the effectiveness of the model.

关 键 词：多特征融合 FISHER准则 图像分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]