基于探索性数据分析的柑橘部位颜色模型分析与识别

Color Model Analysis and Recognition for Parts of Citrus Based on Exploratory Data Analysis

针对成熟柑橘各部位特点、光照与环境的不确定性及多变性,对柑橘各部位的图像数据进行了探索性数据分析并进行了识别。分析了探索性数据分析方法的特点和基于色彩分通道的柑橘识别原理,给出了柑橘图像数据的探索性数据分析流程图。对采集的柑橘图像分成顺光、正常光、逆光3种不同光照情况,并分别采集其中的柑橘果实与果梗、叶子图像数据,根据这些图像数据生成柑橘果实与果梗、叶子在6种颜色模型下的颜色分量分布箱线图,通过图形启示的数据分析与探索,给出了基于I1I2I3颜色模型的I2分量的柑橘各部位分类识别的视觉模型,分析表明I2分量值为0.3能去除树枝、叶子和草地等复杂背景,实现柑橘果实与背景的分割。以300幅野外环境下采集的、不同光照下的柑橘图像为试验对象,成熟柑橘果实总体识别率达到了98.4%,同时证实果梗与叶子由于颜色的相似性,仅靠颜色特征无法对其进行区分。

Aiming at the characteristics of various parts of mature citrus,uncertainty and variability of light and environment,various parts of citrus image data had been conducted to exploratory data analysis and identify.The characteristics of exploratory data analysis method and the principle of citrus recognition based on color channel were analyzed.The flow chart of exploratory data analysis of citrus image data was provided.The acquired citrus images were divided into three different lighting conditions: front-lighting,normal light and backlighting.The image data of citrus fruits, stems and leaves were collected.According to the image data,the sorted box-plot on color component for all parts of citrus based on six kinds of color space were designed.With data analysis about the graphics of box-plot,a vision model of recognition of different parts of citrus was given based on I2color component of I1I2I3color space.When the threshold value of I2 was 0.3,the branches,the leaves and the grass in complex background could be removed and thus citrus fruits and their background could be segregated.Finally,300 differently illuminated citrus images were collected in natural circumstance as test objects,all rip citrus fruits were effectively recognized based on the vision model of I2color feature,and the citrus fruits recognition ratio was 98.4%.Also the situation was confirmed that the fruit stems and leaves could not be distinguished only relying on color feature because of their similarity color.