杂草识别中背景分割方法的比较研究

Research of Different Background Segmentation Algorithms for Weed Identification

应用杂草识别中常用的灰度化方法(非规范化超绿法,归一化超绿法和色差法)针对棉田杂草识别进行了实验。根据灰度图像直方图特点并考虑实时性,采用适当的求阈值法(包括定阈值法和动态阈值法)进行分割,得到5种背景分割方法。通过分割误差对比、各种因素对分割效果的影响对比以及分割的实时性对比,对这5种背景分割方法进行评估,从而为棉田杂草识别中背景分割方法的选取和改进提供依据。

To identify weed in cotton field, the first step is background segmentation. This paper present five background segmentation by combining several grayed algorithms （ excessive green, normalized excessive Green and Cr color differential ） and binarized algorithms properly. First, change the color images into gray images using the grayed algorithms. Then, choose proper binarized algorithms according the characteristics of their histograms. The evaluation and discussion of these background segmentation algorithms can provide references for selection and improving of background segmentation methods in weed detection.