Abstract: The quantitative studies on morphological characteristics of rice root system are of great significance for improving farmland management, rice selective breeding and genetic improvement. In recent years, along with the rapid development of phenotype technology, image processing technology could be used to measure and analyze the growth of rice root system. At the same time, combined with fertilization, irrigation, light, temperature control and environmental monitoring technology, it has become a new technological means for rice breeding and functional genomics research. And the root system image segmentation technology is one of the key foundations for the subsequent phenotypic group analysis. The segmentation is very difficult, because the rice root system images have the characteristics of low contrast, low SNR and complex texture. Aiming at this problem, the paper firstly studied the main-branch connection algorithm, local threshold segmentation algorithm based on the morphological characteristics, and adaptive threshold segmentation based on morphological characteristics, and then performed segmentation and comparison on rice root system images. The results showed that although the main-branch connection algorithm retained a large number of root details, its structure was slightly chaotic and burr phenomenon was serious; the local threshold segmentation algorithm based on morphological features could retain more root details, but the contour fracture was more severe; adaptive threshold segmentation algorithm could improve image continuity, burr phenomenon was controlled, fine roots would not be preserved. These 2 algorithms were combined into a integrated algorithm suitable for image segmentation of rice surface root system. Thus, relatively ideal segmentation results could be obtained, which would provide important basis for follow-up studies on phenotypic traits of rice root system.

Key words: rice root system, phenomics, nondestructive testing, image segmentation

摘要： 水稻根系形态特征的定量研究对于改进农田管理方式、水稻品种选育和遗传改良等具有重要意义。近年来，随着表型组技术迅速发展，利用图像处理技术对水稻根系生长情况进行测量和分析，同时配合施肥、灌溉、光照、温控等环境监控技术已成为水稻育种和功能基因组研究新型技术手段，而根系图像分割技术是进行后续表型组学分析的重要基础之一。由于生长在土壤中的水稻根系图像具有对比度低、信噪比低、纹理复杂的特点，分割十分困难。针对此问题，研究了主干-分支连接算法、基于形态特征的局部阈值分割算法和基于形态特征的自适应阈值分割算法，对生长在土壤中水稻根系图像进行分割处理和比较。实验结果表明，主干-分支连接算法虽然保留了大量细节，但是受噪声影响严重，其结构略显杂乱，毛刺现象严重；基于形态特征的局部阈值分割算法能保留更多根部的细节，但轮廓断裂的现象比较严重；自适应阈值分割算法分割的图像根系连续性较好，毛刺现象也得到了抑制，但是细小的须根无法保留。最终将两种算法结合起来，提出一种适用于水稻表层根系图像分割的综合算法，则可以获得较为理想的分割结果，为后续水稻根系性状提取奠定了重要基础。

关键词: 水稻根系, 表型组学, 无损检测, 图像分割