对光照变化不敏感的微观高光谱图像木材树种识别算法研究
投稿时间:2019-08-18  修订日期:2019-11-19  点此下载全文
引用本文:
摘要点击次数: 107
全文下载次数: 14
作者单位E-mail
王承琨 东北林业大学信息与计算机工程学院黑龙江省哈尔滨市150040 402686820@QQ.COM 
赵鹏 东北林业大学信息与计算机工程学院黑龙江省哈尔滨市150040 bit_zhao@aliyun.com 
基金项目:A. 国家自然科学基金面上项目资助 31670717A. 国家自然科学基金面上项目资助(31670717);“基于显微高光谱成像技术的木材树种分类识别”。
中文摘要:木材往往堆积在室外,在对木材样本采集高光谱图像时往往会受到外界因素(光照、温度、湿度)的影响,从而造成木材树种的误判。为了解决这一问题,本文利用PLS(Pattern Lacunarity Spectrum)和LBP(Local Binary Pattern)对木材横截面的高光谱图像的纹理信息进行了特征提取,而后将高光谱图像的近红外光谱与纹理特征相融合,并以融合后的新特征作为识别的依据,最后使用SVM(Support Vector Machine)和BP(Back Propagation)神经网络两种分类器对木材树种进行了识别,实验表明该算法在无干扰情况下可拥有最高100%的识别正确率效果。为了验证该算法可以在高光谱图像失真的情况下依然可以对木材进行正确的识别,本文仿真了光照变化对高光谱图像的影响,并对比了影响前后的识别正确率,结果显示该算法可以在高光谱图像失真的情况下对木材的树种进行正确的识别,优于传统的和近期主流的木材树种分类算法。
中文关键词:高光谱图像  木材树种识别  光照变化  特征融合
 
Wood species recognition using hyper-spectral images not sensitive to illumination variation
Abstract:Wood is usually stored outdoors so that when its hyper-spectral image is picked up, the acquired image is usually disturbed by environmental factors such as illumination, temperature, and humidity. This disturbance may produce the false wood species classification results. To solve this issue, the wood texture feature is extracted in its hyper-spectral image by use of PLS and LBP. This texture feature is then combined with the near infrared spectra of wood hyper-spectral image so that the fused features are sent into SVM and BP neural network classifiers. Experimental results indicate that our scheme can reach to 100% classification accuracy without environmental disturbance. Moreover, to testify our scheme’s robustness in case of illumination variation, a simulation experiment is performed and it indicates that our scheme outperforms the conventional and the state-of-art wood recognition schemes.
keywords:hyper-spectral image  wood species recognition  illumination variation  feature fusion.
  HTML  查看/发表评论  下载PDF阅读器

版权所有:《红外与毫米波学报》编辑部