面向遥感图像的小样本目标检测改进算法研究  

Improved approach to detect small sample target based on remote sensing image

作　　者：李成范 赵俊娟[2] LI Chengfan;ZHAO Junjuan(Key Laboratory for Digital Land and Resources of Jiangxi Province,East China University of Technology,Nanchang 330013,Jiangxi,China;School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China)

机构地区：[1]东华理工大学江西省数字国土重点实验室,江西南昌330013 [2]上海大学计算机工程与科学学院,上海200444

出　　处：《上海大学学报（自然科学版）》2022年第2期314-323,共10页Journal of Shanghai University:Natural Science Edition

基　　金：东华理工大学江西省数字国土重点实验室开放研究基金资助项目(DLLJ202103)。

摘　　要：针对传统的遥感图像目标检测中面临的小样本以及目标样本分布不均衡等问题,提出了一种基于改进的卷积神经网络(convolutional neural network,CNN)的遥感图像小样本目标检测算法.首先,该算法利用K近邻(K-nearest neighbor,kNN)回归分别对每个点和卷积层提取特征构建局部邻域;同时,通过最大池化聚合所有局部特征进行全局特征表示;最后,采用全连接层与缩放指数型线性单元(scaled expected linear unit,SELU)激活函数计算各类别对应的概率并分类.实验结果表明,该算法能够更有效地融合局部特征,提高了遥感图像小样本目标识别与检测的精度,同时保持信息的非局部扩散.In this study,an improved convolutional neural network(CNN)approach is proposed to detect small sample targets from remote sensing images.The approach is designed to address the two issues of small target samples and the unbalanced distribution of ground object samples with respect to target detection of small samples by remote sensors.In the proposed method,first,K-nearest neighbor(kNN)regression is adopted to extract the features of each point and convolution layer to construct the local neighborhood.Second,all local features are aggregated by maximum pooling layer in CNN to represent global features.Subsequently,the full connection layer and scaled exponential linear unit(SELU)activation function are applied to calculate the probability corresponding to each category for classification.Finally,the proposed approach is tested and evaluated on hyperspectral imager remote sensing images datasets.Experimental results show that the proposed improvements to the CNN model fuse fully local features and result in the effective recognition and detection of small sample targets from remote sensing image with high accuracy while maintaining the nonlocal diffusion capabilities of information.

关 键 词：卷积神经网络 小样本 深度学习 

分 类 号：TB567[交通运输工程—水声工程]