结合局部特征的遥感飞机目标识别  

Target Recognition of Remote Sensing Aircraft Combining with Local Features

作　　者：于丽 YU Li(College of Information Engineering,Shanghai Maritime University,Shanghai 201306)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《现代计算机（中旬刊）》2018年第8期53-57,71,共6页Modern Computer

基　　金：国家自然科学基金(No.61271446);航空科学基金(No.2013ZC15005)

摘　　要：在被遮挡的遥感飞机图像中,飞机目标特征由于遮挡而部分缺失,导致卷积神经网络提取不到充足的特征信息进行分类,识别准确率不高。卷积神经网络的缺点之一就是池化层的存在会使很多有价值的信息被滤除,也会忽略全局信息和局部信息之间的关联,这给遮挡图像识别造成很大影响。因此针对卷积神经网络对遮挡目标识别率低的问题,将局部特征(尺度不变特征)与卷积特征结合,以增加关键特征信息,提高算法对于遮挡情况的判别力。In the occluded remote sensing aircraft image, the aircraft target feature is partially lost due to occlusion, resulting in the con-volutional neural network extracting insufficient feature information for classification, and the recognition accuracy rate islow. One of the disadvantages of convolutional neural networks is that the presence of the pooling layer can filter out manyvaluable information and ignore the correlation between global information and local information, which has a great impact onocclusion image recognition. Therefore, for the problem that the convolutional neural network has a low recognition rate for oc-clusion targets, local features（scale-invariant transform features）and convolution features are combined to increase key fea-ture information and improve the discriminative power of the algorithm for occlusion conditions.

关 键 词：遥感图像 目标识别 尺度不变特征 密集卷积特征 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]