高级检索
    陈珠琳, 王雪峰. 檀香咖啡豹蠹蛾虫害的树干区域分类研究[J]. 北京林业大学学报, 2018, 40(1): 74-82. DOI: 10.13332/j.1000-1522.20170306
    引用本文: 陈珠琳, 王雪峰. 檀香咖啡豹蠹蛾虫害的树干区域分类研究[J]. 北京林业大学学报, 2018, 40(1): 74-82. DOI: 10.13332/j.1000-1522.20170306
    Chen Zhu-lin, Wang Xue-feng. Classification of sandalwood trunk area damaged by Zeuzera coffeae nietner in complex background[J]. Journal of Beijing Forestry University, 2018, 40(1): 74-82. DOI: 10.13332/j.1000-1522.20170306
    Citation: Chen Zhu-lin, Wang Xue-feng. Classification of sandalwood trunk area damaged by Zeuzera coffeae nietner in complex background[J]. Journal of Beijing Forestry University, 2018, 40(1): 74-82. DOI: 10.13332/j.1000-1522.20170306

    檀香咖啡豹蠹蛾虫害的树干区域分类研究

    Classification of sandalwood trunk area damaged by Zeuzera coffeae nietner in complex background

    • 摘要:
      目的檀香是一种典型的珍贵树种,但不易成活且常遭受病虫害的干扰。本研究针对受咖啡豹蠹蛾危害的檀香,为提高受害檀香树干区域的分类效果,根据不同区域在颜色和纹理方面的特点,提出了一种复杂背景下受咖啡豹蠹蛾危害的檀香树干区域分类方法。
      方法(1) 本文首先结合高斯高通滤波、Otsu法、2G-R-B因子以及形态学运算,将树干从复杂背景中提取出来,再通过L*a*b*系统中的L*通道和a*通道将树干分割为健康区域、虫害区域和排泄物区域。(2)提出“多纹理”并使用“相对颜色”特征变换方法以提取图像特征。(3)提出BP神经网络和RBF-SVM相结合的方法对区域进行识别。
      结果(1) 本文提出的图像分割算法有效地将受虫害危害的檀香从背景中提取出来,并成功分割为健康区域、虫害区域和排泄物区域。(2)与传统PCA处理相比,使用“多纹理”和“相对颜色特征”得到的分类精度有所提高。其中,“多纹理”特征通过差异扩大的方法构建出的图像特征,明显增加了类间方差;“相对颜色特征”则减小了光照对样本的影响,从而减小了类内方差。(3)通过将RBF-SVM与BP神经网络方法相结合,对比发现,使用一种方法进行三分类的精度较低,分别为74.44%和81.11%;两次二分类后提高了精度,分别为87.77%和85.56%;而两种方法相结合的方式所得到的分类精度最高,总体可达91.11%。
      结论本文通过数字图像对受咖啡豹蠹蛾危害的檀香树干区域进行分类,为该虫害的早期识别以及受损率计算提供了方法,可以有效地减小檀香生长所受危害,提高心材质量,也为数字图像技术在林业中的应用提供了参考。

       

      Abstract:
      ObjectiveSandalwood is one of the precious tree species which is hard to survive and often suffers from diseases and pests. In order to improve the classification accuracy of sandalwood trunk regions, this research proposed a classification method for different sandalwood trunk regions which is attacked by coffee carpenter moth.
      Method(1) First of all, we combined Gauss high-pass filter, Otsu segmentation method, 2G-R-B factor with morphological operation together to extract the sandalwood from background and divided the trunk into three regions, i.e. healthy region, pest region and excrement region using L* channel and a* channel in L*a*b* system. (2) "Multi-texture" was proposed and "relative color" was used to extract image features. (3) We combined RBF-SVM and BP neural network together to identify different regions.
      Result(1) The image segmentation algorithm proposed in this paper extracted the sandalwood from background and successfully divided it into healthy region, insect region and excrement region. (2) Compared with the traditional PCA processing, the classification accuracy obtained using "multi-texture" and "relative color" features was improved. Among them, the "multi texture" features were constructed by enlarging the difference of image features, which significantly increased the variance between classes. The "relative color" features reduced the influence of illumination on samples, thus reduced the intra class variance. (3) We compared different classification methods and found that only using RBF-SVM or BP neural network often caused low accuracy, the results were 74.44% and 81.11%, respectively. After using binary classification twice, the results were improved to 87.77% and 85.56%, respectively. The best classification result was obtained by combining RBF-SVM and BP neural network together, which was 91.11%.
      ConclusionThis paper provides a new classification method for sandalwood trunk regions which is attacked by coffee carpenter moth by digital image processing. It provides a method for early identification and damage rate calculation of insect pests. It can reduce the harm caused by pests, improve the quality of heartwood, and provide a reference for the application of digital image technology in forestry.

       

    /

    返回文章
    返回