基于非线性尺度空间的航拍场景分类

发布时间：2018-12-20 12:06

【摘要】：针对尺度不变特征变换(Scale Invariant Feature Transform,SIFT)算法在航拍场景分类中提取特征时,易造成边界模糊和细节丢失且无法描述颜色信息的问题,结合视觉词袋模型,提出了非线性尺度空间下融合颜色特征的新型颜色风式特征检测子(Color-KAZE,C-KAZE).通过KAZE构造非线性尺度空间来检测特征信息;对颜色模型(Hue,Saturation,Value,HSV)非等间隔量化获取颜色量化矩阵,进而生成C-KAZE特征描述子;利用视觉词袋和空间金字塔匹配模型融合多特征.实验表明,该算法相比SIFT算法在场景分类准确率方面提高了约8%.C-KAZE描述子增强了KAZE的特征描述能力,突破了SIFT算法特征描述单一、边缘细节模糊的局限性,显著提升了无人机航拍图像的分类效果.
[Abstract]:Aiming at the problem of feature extraction in aerial scene classification by scale invariant feature transform (Scale Invariant Feature Transform,SIFT) algorithm, it is easy to cause boundary blur and detail loss and can not describe color information. A novel color wind-style feature detector (Color-KAZE,C-KAZE) with color features fusion in nonlinear scale space is proposed. The nonlinear scale space is constructed by KAZE to detect the feature information, the color quantization matrix is obtained by non-equal-interval quantization of the color model (Hue,Saturation,Value,HSV), and the C-KAZE feature descriptor is generated. Visual word bag and spatial pyramid matching model are used to fuse multiple features. Experiments show that compared with SIFT algorithm, the algorithm improves the accuracy of scene classification by using the approximate 8%.C-KAZE descriptor to enhance the feature description ability of KAZE, and breaks through the limitations of the SIFT algorithm that the feature description is single and the edge details are fuzzy. The classification effect of aerial images of UAV is improved significantly.
【作者单位】：南京信息工程大学江苏省气象探测与信息处理重点实验室;
【基金】：国家自然科学基金(61302188) 中国博士后特别资助基金(2012T50510,2011M500940) 六大人才高峰项目(2013-DZXX-020) 江苏省高校重大自然科学基金项目(12KJA510001)资助
【分类号】：TP391.41

【相似文献】