基于探索性分析的的荔枝果及结果母枝颜色特征分析与识别

Color feature analysis and recognition for litchi fruits and their main fruit bearing branch based on exploratory analysis

从复杂背景中识别成熟荔枝串中的荔枝果及结果母枝,获取结果母枝上的采摘点是机器人视觉定位与识别的难点,荔枝果、结果母枝与叶子各部位图像颜色特征分析与识别成为研究重点。首先针对荔枝果与结果母枝的特点、光照与环境的特殊性及不确定性,提出了探索性分析与荔枝图像识别的融合方法,对荔枝果与结果母枝进行了图像分类与统计的探索性分析,并给出了荔枝图像数据的探索性分析流程图;其次,根据荔枝不同部位颜色均值分布的特点,设计了荔枝果、结果母枝及叶子在6种色彩模型下的颜色均值分布箱线图,通过图形启示的数据分析与探索,给出了基于YCbCr色彩空间的Cr单通道图的荔枝各部位分类识别的视觉模型,分析表明Cr分量值在0.5～0.54能去除叶子和侧枝等复杂背景,实现荔枝串中的荔枝果与结果母枝的分割。最后,以60组不同光照条件的180幅自然环境下采集的荔枝图像为试验测试对象,用颜色特征的视觉模型结合阈值分割方法有效地识别了成熟荔枝串与荔枝果,荔枝串与荔枝果的平均识别率分别为91.67%和95.00%。用探索性分析与图像运算相结合的方法成功地提取了结果母枝(识别率为86.67%),并用计算出的采摘点进行视觉定位的仿真。试验和仿真结果表明视觉模型及其方法能对荔枝不同部位进行有效识别。

For harvesting robots, it is difficult to recognize and locate ripe litchi fruits and their main fruit bearing branch from clusters of litchi in complicated background. Hence, analyzing and recognizing color feature of the images of litchi fruits, litchi main peduncles and leaves become research focuses. Firstly, according to the specialty and uncertainty of litchi fruits, litchi main fruit bearing branch, the method of combining the exploratory analysis with litchi image recognition was put forward in this research by discovering the uncertain element spaces of targets in images, sorting the litchi main fruit bearing branch into three colors of partial green, partial red and partial brown, sorting the litchi fruits-influencing illumination into highlight, normal light and backlighting, thus classifying and analyzing the images of all the parts of litchi, and then providing the flow chart of the exploratory analysis of litchi image data. Secondly, the exploratory analysis about color feature of litchi fruits, litchi main fruit bearing branch and leaves was illustrated in this research, and sorted box-plot on color component for all parts of litchi based on the RGB（red, green, blue）, HSI（hue, saturation, intensity ）, Lab （L stands for lightness, and a and b stand for the color-opponent dimensions based on nonlinearly compressed CIE XYZ color space coordinates）, YCbCr （luminance is denoted by Y, Cb and Cr are the blue- difference and red-difference）, normalized rgb and I11213 were designed. With data analysis about the graphics of box-plot, a vision model of recognition of different parts of litchis was given based on Cr gray-scale image of YCbCr color space. When the threshold value of Cr was between 0.50 and 0.54, the leaves and lateral branches in complex background can be removed and thus litchi fruits and their main fruit bearing branch from the cluster of litchi can be segregate. Finally, taking 60 groups （all together 180） of differently illuminated litchi images collected in natural circumstance as test objects, all ripe litchi clusters and litchi fruits of testing images with the threshold segmentation method were effectively recognized based on the vision model of Cr color feature, and their recognition ratio was 91.67% and 95.00%, respectively. After that, the main fruit bearing branch was successfully extracted from the recognized litchi cluster by operation on the recognized images （with recognition ratio 86.67%）, and with the calculated picking-point, the visual location simulation was carried out. The results of the test experiment and visual simulation attest that vision model based on Cr gray-scale image of YCbCr color space combined with the corresponding segmentation method can effectively recognize different parts of litchi.