Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (2): 177-181.DOI: 10.3778/j.issn.1002-8331.1608-0084

Previous Articles     Next Articles

Research on farmland image segmentation based on SLIC method under strong light

CHEN Xiaoqian, TANG Jinglei, WANG Dong   

  1. College of Information Engineering, Northwest A&F University, Yangling, Shaanxi 712100, China
  • Online:2018-01-15 Published:2018-01-31


陈晓倩,唐晶磊,王  栋   

  1. 西北农林科技大学 信息工程学院,陕西 杨凌 712100

Abstract: Precision agriculture is the trend of agricultural development, while farmland image segmentation is the pre-mise and foundation of precision agriculture. Since the effect of farmland images highlights areas lost plant features on the quality of the images segmentation, based on the SLIC method and Cg component in YCrCb color space, taking use of different classifiers, this paper realizes farmland images segmentation under the condition of strong light. Firstly, farmland images are pre-processing using SLIC, to obtain super pixel region. To avoid plant pages losing green features because of the highlight areas under strong light, this paper introduces the Cg component in YCrCb color space and excess green characteristic. To avoid high requirements for the training sample in supervised learning, this paper uses semi-supervised learning, mixes up the labeled samples with samples without a label. Finally, images segmentation takes different classifiers, and the quality of image segmentation is evaluated by confusion matrix and Kappa coefficient. Contrasting the experimental results, the kernel function which is diagQuadratic in distance discriminance is better than other methods, the accuracy is higher.

Key words: image segmentation, different classifiers, Simple Linear Iterative Clusterign(SLIC) method, Cg component, strong light

摘要: 精准农业是未来农业发展的趋势,而农田图像分割是精准农业的前提与基础。针对光照偏强条件下农田图像高光点区域丢失植物绿色特征对图像分割质量的影响,以SLIC方法和YCrCb颜色空间中的Cg分量为基础,利用不同分类器实现光照偏强条件下农田图像分割的研究。首先采用SLIC对农田图像进行预处理,获取超像素模块;为避免植物叶面因光照偏强出现高光点区域丢失部分绿色特征,引入YCrCb颜色空间模型中的Cg分量和超绿颜色因子提取特征;为避免监督学习对训练样本要求高,采用半监督学习,将有标签样本和无标签样本进行混合;最后采用不同的分类器进行图像分割,并对实验结果采用混淆矩阵和Kappa系数进行评价。对比实验结果可得,采用距离判别法核函数为diagQuadratic的图像分割效果较其他方法较好,分割正确率较高。

关键词: 图像分割, 不同分类器, 简单的线性迭代聚类(SLIC)方法, Cg分量, 光照偏强