基于K-means聚类和图像分割的紫色土发生层边界识别

Boundary Identification of Purple Soil Horizon Based on K-means Clustering and Image Segmentation

土壤学始于对土壤剖面及其形态特征的观察,剖面发生层的划分与发生层边界特征的描述是土壤调查的基础。实地划分发生层需要丰富的土壤学实践经验,存在主观和缺乏统一划分标准的问题。以紫色土剖面图像为研究对象,采用K-means聚类和图像分割技术,结合图像的颜色特征(CIELab色彩空间)和纹理特征(Entropy)识别紫色土剖面发生层边界,并与实地划分的结果进行比较。结果表明:(1)CIELab色彩空间的a、b通道和Entropy纹理特征,可以划分出供试剖面的主要发生层(A、B、C)和基岩(R);(2)聚类识别的发生层数量和发生层深度与实地识别的结果基本一致;除Z2剖面的C层和Z6剖面的Ap层聚类识别与实地识别的发生层下边界深度相差较大(分别为13cm和8cm)外,其余发生层下边界深度相差均在3 cm以内;(3)聚类识别的发生层边界形状更为不规则,明显度更为模糊。K-means聚类和图像分割技术实现了紫色土剖面发生层边界的客观识别,可为土壤剖面智能辨识系统的开发提供科学参考。

【Objective】Pedology begins with the observation of soil profile and its morphological characteristics.The division of the soil profile horizon and description of the characteristics of the horizon boundary are the basis of soil investigation.The division of soil horizon in the field requires rich practical experience in pedology and is more subjective,which makes it difficult to form a set of unified division standards.【Method】In this paper,the purple soil profile image was taken as the research object,and using K-means clustering and image segmentation technology,combined with the color(CIE Lab color space)and texture characteristics(Entropy)of the image,we identified the horizon boundary of the purple soil profile,by comparing with the results of field division.【Result】The results show that(1)the a and b channels of CIE Lab color space and Entropy texture characteristics can delineate the master horizon(A,B,and C)and bedrock(R)of the profile;the a channel values range from 7-22,the b channel values range from 7-19,and the Entropy values were 4 or 5;the Munsell colors converted by the CEL XYZ system had a certain deviation from the colors visually discerned in the field using colorimetric cards,with a hue range of 10R-2.5Y,a value range of 4-8,and a chroma range of 3-8.(2)The number of soil horizon and the depth of soil horizon identified by clustering were consistent with the results of field identification;the difference between the lower boundary depth of soil horizon identified by clustering identification and those identified in the field was within 3 cm,except for C in profile Z2 and the Ap in profile Z6,where the difference was larger(13 cm and 8 cm,respectively).(3)The topography of the soil horizon identified by clustering was more irregular and the distinctness was more blurred.The clustering algorithm can identify more subtle differences in the soil profile image and reflect the local variation of soil properties in more detail.【Conclusion】K-means clustering and image segmentation techniques achieved the identification of the horizon boundary of purple soil,and this study provides a scientific reference for the development of an intelligent identification system for soil profiles.