基于灰度共生矩阵与SOM神经网络的树皮纹理特征识别

Identification of Tree Bark Texture Characteristic Based on Gray Co-occurrence Matrix and SOM Neural Network

黄檗、水曲柳、胡桃楸3种林木的树皮提取物作为名贵中药的植物来源。通过环剥再生技术采集树皮,在保证经济需要的同时有效地保护了林木。但3种林木树皮纹理相近且不易区分,现场采集给林业工作者带来困难。因此通过图像处理技术解决这一难题具有实际意义。采集3种林木树皮图像各300幅,共计900幅,对图像分别进行ROI(感兴趣区域)截取、直方图均衡,构造d=2;g=128;θ=0°、45°、90°、135°的灰度共生矩阵,提取14个特征参数;通过数字特征分析,筛选出8个特征值;应用SOM(Self Organizing Maps)神经网络对大量林区树皮图像进行参数验证。得到由角二阶矩、熵、相关性、方差、聚类阴影、和熵、聚类阴影构成的参数集有效,识别精度83.33%。证明该方法可以很好地区分黄檗、水曲柳、胡桃楸3种林木。

The extractives of tree bark of Cortex Phellodendri 9 Fraxirm s mand shurica and Catalpa are used as the important source of precious Chinese medicine. The bark is harvested by girdling regeneration technology which ensures the economic needs and also protects the trees. Due to the tree barks of three species are too similar to easily distinguish, it brings the difficult to forestry workers. Therefore it is of practical significance that uses the image processing technology to solve this issue. A total 900 tree bark im-ages of three species, 300 per species, were collected. ROI（Region of Interest）image capture and histogram equalization were con-ducted. The gray level co - occurrence matrix with c？ of 2, g of 128 and the 0 of 0° , 45° , 90°and 135 ° were constructed. 14 char-acteristic parameters were extracted. 8 characteristic parameters were selected effectively through the analysis of digital characteristics. The SOM （Self Oi^anizing Maps） neural network was used to conduct parameters validation for a lai^e number of tree bark images. The parameters set consisting of angular second moment, entropy, moment of inertia, correlation, variance, clustering shadow, and en-tropy can be used to effectively distinguish three tree species with the recognition of 83. 33%. The method studied in this paper is ca-pable to well distinguish three trees species of Cortex Phellodendri, Fraxinus mandshurica and Catalpa .