基于K-means聚类与果萼形状的‘海沃德’猕猴桃膨大果检测方法

Identifying expanded‘Hayward’kiwifruits based on K-means clustering algorithm and calyx shape

【目的】根据‘海沃德’猕猴桃膨大果和未经膨大剂处理的猕猴桃正常果在果萼形状上的差异,建立一种基于K-means聚类算法的猕猴桃膨大果的图像识别方法。【方法】利用K-means聚类算法对猕猴桃的RGB图像进行聚类,以初步分割出猕猴桃果萼;对RGB图像进行灰度化,并采用最大类间方差法对灰度图像进行阈值分割,基于阈值分割后的二值图像提取猕猴桃的边缘区域;基于K-means聚类分割的结果与猕猴桃边缘区域的二值图像提取果萼的聚类,再利用数学形态学处理精确提取出猕猴桃的果萼。求取果萼区域最小外接矩形的长宽比,并根据长宽比进行猕猴桃正常果与膨大果的判断与识别。【结果】对提取的猕猴桃果萼特征的统计分析表明,当猕猴桃果萼区域最小外接矩形的长宽比值大于1.6时为膨大果,否则为正常果,利用该算法对‘海沃德’猕猴桃膨大果的总体正确识别率为91.5%。【结论】基于K-means聚类分割算法和果萼形状上的差异所提出的猕猴桃膨大果无损、便捷检测方法,为猕猴桃膨大果的工业化在线检测及基于手机平台的猕猴桃膨大果识别软件的开发提供了新思路。

【Objective】The objective of this study was to provide a novel method for the identification of expanded‘Hayward’kiwifruits based on K-means clustering algorithm and the differences in calyx shape.【Method】K-means clustering algorithm was applied to RGB images of kiwifruit to segment the calyx initially.The RGB images were grayed before the Otsu algorithm was used for threshold segmentation.The edge region of kiwifruit was extracted based on segmented images.Based on the clustering result and binary images of edge region,the cluster containing calyx region was extracted only and the calyx was accurately extracted using mathematical morphology processing.The length-width ratio of the minimum enclosing rectangle of calyx was calculated,and kiwifruits were identified based on it.【Result】Based on the statistics of extracted calyxes,expanded kiwifruits were identified when the length-width ratio of the minimum enclosing rectangle of calyx was higher than 1.6.Otherwise,the kiwifruit was regarded as a normal one.The total identification accuracy of the method was 91.5%.【Conclusion】The non-destructive,quick and cheap method for identification of expanded kiwifruits based on K-means clustering algorithm and calyx shape was established successfully.It provides a new idea for industrial on-line detection of expanded kiwifruits and the development of software to identify expanded kiwifruits based on smartphone platform.