显著图和多特征结合的遥感图像飞机目标识别

发布时间：2018-05-21 02:10

本文选题：飞机目标识别 + 遥感图像　；参考：《中国图象图形学报》2017年04期

【摘要】：目的遥感图像飞机目标的检测与识别是近年来国内外研究的热点之一。传统的飞机目标识别算法一般是先通过目标分割,然后提取不变特征进行训练来完成目标的识别。在干扰较少的情况下,传统算法的识别效果较好。但遥感图像存在着大量的干扰因素,如光照变化、复杂背景及噪声等,因此传统算法识别精度较低,耗时量较大。为快速、准确识别遥感图像中飞机目标,提出一种基于显著图和全局特征、局部特征结合的飞机目标识别算法。方法首先使用改进的Itti显著算法提取遥感图像中的显著目标;接着使用基于区域增长和线标记算法寻找连通区域来确定候选目标的数量和位置;然后提取MSA(multi-scale autoconvolution)、Pseudo-Zernike矩和HarrisLaplace特征描述子,并使用标准差和均值的比值来评估特征的稳定性,再把提取的特征结合成特征向量;最后应用支持向量机的方法完成对候选目标的识别。结果实验结果表明,本文算法检测率和识别率分别为97.2%和94.9%,均高于现有算法,并且耗时少,虚警率低(为0.03),对噪声干扰、背景影响以及光照变化和仿射变化均具有良好的鲁棒性。结论本文算法使用了图像的3种特征信息,包括MSA、Pseudo-Zernike矩和Harris-Laplace特征描述子,有效克服单一特征的缺点,提高了遥感图像飞机目标的识别率和抗干扰能力。
[Abstract]:Objective the detection and recognition of aircraft targets in remote sensing images is one of the hot research topics at home and abroad in recent years. The traditional aircraft target recognition algorithm usually uses target segmentation, and then extracts invariant features for training to complete target recognition. In the case of less interference, the recognition effect of the traditional algorithm is better. However, there are a lot of interference factors in remote sensing image, such as illumination variation, complex background and noise, etc. In order to identify aircraft targets in remote sensing images quickly and accurately, an algorithm of aircraft target recognition based on salient map and global features and local features is proposed. Methods first, the improved Itti saliency algorithm is used to extract salient targets in remote sensing images, and then the number and location of candidate targets are determined by searching connected regions based on region growth and line marking algorithms. Then MSA(multi-scale autoconvolutional pseudo-Zernike moments and HarrisLaplace feature descriptors are extracted, the stability of the features is evaluated by the ratio of standard deviation and mean value, and the extracted features are combined into feature vectors. Finally, support vector machine is used to complete the recognition of candidate targets. Results the experimental results show that the detection rate and recognition rate of the proposed algorithm are 97.2% and 94.9%, respectively, which are higher than those of the existing algorithms, and take less time, and the false alarm rate is low (0.03%), which interferes with the noise. Background effects, illumination changes and affine changes have good robustness. Conclusion the algorithm uses three kinds of feature information of image, including MSA-Pseudo-Zernike moment and Harris-Laplace feature descriptor, which overcomes the shortcoming of single feature and improves the recognition rate and anti-jamming ability of aircraft target in remote sensing image.
【作者单位】：南昌航空大学计算机视觉研究所;
【基金】：国家自然科学基金项目(61165011,61662049) 南昌航空大学研究生创新专项资金项目(YC2015058)~~
【分类号】：TP751

【参考文献】