桃子表面缺陷分水岭分割方法研究  

Watershed Segmentation Method for Segmenting Defects on Peach Fruit Surface

作　　者：李江波[1,2] 彭彦昆[1] 黄文倩[2] 张保华[2] 武继涛[2] 

机构地区：[1]中国农业大学工学院,北京100083 [2]北京农业信息技术研究中心,北京100097

出　　处：《农业机械学报》2014年第8期288-293,共6页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金资助项目(31301236);北京市博士后科研活动经费资助项目(2013ZZ-70);2012年北京市农林科学院博士后基金资助项目

摘　　要：水果表皮缺陷的有效检测是水果自动化无损检测重要的部分,并且表皮缺陷的准确分割是对缺陷果准确分级的前提,也有助于缺陷果的分类识别。然而,通常水果表面具有较大的曲率变化,这种变化的曲率导致水果表面对同一入射光源照度反射的不均,进而影响缺陷区域的准确分割。本研究以平谷大桃为例,提出采用基于形态学梯度重构和内外标记的分水岭算法对水果表面缺陷进行分割。首先,提取R通道图像,采用NIR图像构建掩模模板并对R通道图像去背景;随后,去除背景后的图像进行形态学梯度变化获取梯度图像,并对梯度变化后的图像进行梯度重建以去除水果表面的细小噪声;接着,对重建后的梯度图像进行形态学标记运算获取标记图像;然后,采用标准分水岭算法实现缺陷的准确分割。对正常果、刺伤果、裂果、黑斑果、虫咬伤果、腐烂果和疤伤果7种表皮类型样本共计525幅图像检测结果表明,能够获得96.8%识别率。实验证明,基于形态学梯度重构和标记提取的分水岭算法能够有效用于桃子表面缺陷的分割,并不会受到桃子表面光照不均的影响。Effective detection of destructive detection of fruit. An grade the fruit based on size of peel defects on fruit was always the most important in automatic non- d, accurate segmentation of peel defects was a premise to effectively defect. However, since fruit surface usually has a larger curvature change, the non-uniform reflection from fruit surface is probably caused in terms of the same incident light source, and the accurate segmentation of peel defects will be affected. ＇ Pinggu＇ peaches were applied as the research object, and a watershed segmentation method combining morphological gradient reconstruction with internal and external markers was proposed to segment the defects on fruit peel. First, R channel image was extracted and background was removed by mask template obtained from NIR image. Then, gradient image was obtained by morphological gradient operation, and gradient reconstruction was performed by using the gradient image to remove some small noises on the fruit surface. Next, internal and external marker operations were used to obtain the marker image. Finally, defects on peel were segmented by using the standard watershed algorithm. For the investigated 525 sample images including seven peel types, a 96. 8% successful recognition rate was achieved. The experimental results showed that a watershed segmentation algorithm combining morphological gradient reconstruction with marker extraction could be effectively used to segment the peel defects on peach and the performance of algorithm was not affected by non-uniform illumination on peach surface. However, defect segmentation rate needed to be further improved.

关 键 词：桃子 缺陷分割 分水岭算法 形态学变化 标记提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]