摘要
在外科神经修复手术中,正确识别神经束性质是实现良好修复的关键。本文将显微高光谱成像技术应用于神经分类中,并对分类的可行性进行实验性探究。首先使用显微高光谱神经采集系统获取兔子运动及感觉神经的高光谱图像数据并进行预处理,再利用纯净像元提取算法得到端元波谱继而获取各类别的特征光谱,通过分析特征光谱的特征与差异找寻合适的分类算法,实验结果表明本技术具有一定的分类效果。本文基于K近邻分类器,实验性的使用了经典欧氏距离及波谱角距离这两种距离测度算法对实验数据进行分类,对比实验结果分析两种方法的优劣,为后续寻找其他合适且更具针对性的分类方法奠定了重要的基础。
In the nerve repair surgery,the correct identification of nerve fascicles is the key to achieve good repairing result.This article will apply microscopic hyperspectral imaging technology in nerve classification and do some experimental researches on the feasibility of nerve classification.The results showed that it had certain effect.First,the authors uses microscopic hyperspectral neural collecting system to get the rabbit＇s motor and sensory nerves＇ hyperspectral image data,and then did pre-processing.Second,we obtained endmembers by using the pure endmember extraction algorithm and got each category＇s characteristic spectrum.Last,we analyzed the characteristics and differences of spectrum to find suitable classification algorithms.Based on Knearest neighbor classifier,the authors used two classic and simple distance measure algorithms involve euclidean distance and spectral angle match distance to do classification in the experimental data.The results gotten by analying the advantages and disadvantages of the two methods can lay an important foundation for seeking other appropriate and more targeted classification methods.
出处
《影像科学与光化学》
CAS
CSCD
北大核心
2015年第3期203-210,共8页
Imaging Science and Photochemistry
基金
国家自然科学基金项目(61240006)资助
关键词
显微高光谱
神经分类
端元
特征光谱
K近邻分类
microscopic hyperspectral imaging
nerve classification
endmember
characteristic spectrum
K-NN classification
