数字图像处理技术结合偏最小二乘法定量分析柑橘质量

Partial Least Square (PLS) Quantitative Analysis of Citrus Fruit Weight Based on Digital Image Processing

目的:构建柑橘质量模型,实现其计算机视觉法质量检测与在线分级。方法:以湖南湘中成熟中期涟红温州蜜柑为研究对象,对计算机视觉系统采集的各柑橘的4幅图像进行图像裁切、图像去背景、图像二值化与取反操作后,提取柑橘区域像素,以柑橘区域像素与图像像素比为自变量,建立柑橘质量偏最小二乘回归(PLS)预测模型。结果:柑橘质量预测绝对误差为-14.9092～4.9981g,平均误差为-3.9278g,误差标准差4.5210g,±10g质量内的正确识别率为93.3333%,±8g质量内正确识别率76.6667%。结论:通过计算机视觉技术进行柑橘质量在线分级和生长期中挂果质量的监测是可行的。

A model for citrus weight measuring was developed to realize online grading of citrus fruit with computer vision system. Four images from each of mid-maturing Lianhong Wenzhou citrus fruits （grown in the middle part of Hunan province） captured by computer vision system were processed by cutting, background removal, binary conversion and negative operation and finally the images of fruit part were extracted. The pixel ratios of fruit part/total image were used to construct partial least square （PLS） regression model to estimate fruit weight. Results showed that the absolute Weight estimation error was -14.9092 -4.9981 g and the average error was -3.9278 g. The standard deviation was 4.5210 g, and the correctness for accuracy ±10 g was 93.3333% and for accuracy ± 8 g 76.6667%. It is feasible to estimate citrus fruit weight online and monitor the fruit weight during growing period by computer vision system.