檀香树干受咖啡豹蠹蛾危害区域的分割与分类

为了实现海南省北部县市檀香受咖啡豹蠹蛾(Zeuzera coffeae Nietner)虫害的自动识别,使用林内传感器传回的图像信息,提出一种空域与频域相结合的背景去除方法,并提取出虫害区域与健康区域。该方法首先提取出檀香树的前景部分,使用2G-B-R因子去除枝叶及边缘,在L*a*b*系统中选择合适的通道,使用Otus法和形态学运算剔除排泄物区域,并成功分割出虫害和健康区域。通过2种区域的图像在纹理方面表现出的不同,提取筛选出3种受外界因素影响比较小的特征,并在此基础上利用差异扩大法提出了"多纹理特征"的概念。使用Logistic二分类法对提取出的纹理特征及其组合、多纹理特征及其组合、主成分分析后的特征进行分类并分析,结果证明通过扩大差异得到的多纹理特征分类效果要好于单纹理特征,且使用"熵值均值-相关性均值"得到的分类精度最高,并使用系统聚类以及K-means聚类方法验证得到相同的结论,证明了所提方法的科学性。

Segmentation and Classification of Coffee Leopard Moth Attacked Area in Sandalwood Trunk

To discuss image understanding about the case of coffee leopard moth pest on sandalwood in north of Hainan,we proposed a classification method by using images obtained in sensors which put in the forest.We used combined spatial and frequency-domain information to segment trunk from complex background,and used 2G-B-R to clear branches,leaves and bright edges in the foreground.After the trunk extraction was completed,we selected the best channel of L* a* b* system and combined the Otus method and mathematical morphology to segment regions successfully.Three kinds of texture features,which were less affected by external factors,were extracted and the concept of multi-texture features is proposed by means of different expansion method.We used the logistic model classification to classify and analyze the extracted texture features and their combinations,multi texture features and their combinations,principal component analysis.The "entropy mean-correlation mean" had the highest classification accuracy,and the same result was obtained by system clustering and the K-means clustering method.