摘要
为了解决飞机、直升机、导弹等3类空中目标图像的自动分类问题,提出了一种基于改进词袋模型的空中目标识别方法。首先采集3类多个型号的空中目标灰度图像并分割提取出目标,接着利用稠密采样方法进行SIFT特征提取,然后用模糊C均值聚类方法,对空中目标图像的SIFT特征进行聚类,得到大量空中目标图像的视觉单词。最后用视觉单词直方图训练支持向量机分类器,完成空中目标的自动分类。仿真实验表明,文中提出的算法能准确区分空中目标类别,性能优于传统的采用K均值聚类的词袋模型,且优于仿射矩。
In order to solve three types of aerial target （aircraft, helicopter, missile） images automatic classification, an improving bag of words model for recognizing aerial targets was proposed. Firstly, many grey images of three types of aerial targets were collected and segmented for extracting target from background. Secondly, those images, which only included target, were sampled and extracted using scale- invariant feature transform （SIP-T） descriptored by sparse sampling. Thirdly, fuzzy C means （FCM） clustering was used and plenty of visual words on target images were acquired. Finally, histograms of visual words were used to training support vector machine （SVM） classifier and aerial target type would be identified. Experiments are shown that the proposed algorithm can distinguish the types of aerial targets effectively. Moreover, combining with the SVM classifier, the recognition rate by the proposed algorithm is better than traditional ba~ of words model by K means clusterin~ and affine moments.
出处
《红外与激光工程》
EI
CSCD
北大核心
2012年第5期1384-1388,共5页
Infrared and Laser Engineering
基金
国家高技术研究发展计划(863)项目(2010AA7080302)
关键词
空中目标
粗分类
词袋模型
模糊C均值聚类
支持向量机
aerial target
coarse classification
bag of words
FCM clustering
support vector machine
