基于机器视觉与光谱融合的柑橘品质无损检测分级系统设计与试验

Design and experiment of non-destructive testing and grading system for citrus quality based on machine vision and spectral fusion

针对柑橘果径、着色率和内部糖度3项关键品质指标,基于双锥滚子式果杯传输线设计了一套柑橘综合品质无损检测分级系统,该系统主要包括喂料部分、机器视觉检测模块、近红外光谱检测模块和分级执行部分.机器视觉检测模块采用单相机拍摄不断翻滚的柑橘视频来获取大量不同姿态的柑橘图像,并进行轮廓提取,以单个柑橘所有帧图像的最小外接圆直径的平均值计算果径,以每一帧图像得到的其二维黄色占比的平均值作为全表面着色率.在近红外光谱检测模块中设计了透射式光路,采集柑橘透射率光谱,并按在线检测时柑橘出现的两种高频姿态建立了混合姿态糖度检测模型,对比不同预处理方法下的建模结果,选取应用效果较优的多元散射校正(MSC)后建立的偏最小二乘法(PLS)模型.在线试验结果表明:果径检测的最大绝对误差为-1.42 mm,着色率检测的最大绝对误差为0.048,糖度检测结果的相关系数为0.817,均方根误差为0.658%.内外品质的联合检测分级按判别树决策方法确定了3种品质的联合分级方式,在分选速度为5个/s时,综合分级的平均准确率可达到91.16%,该检测分级系统整体结构简单,对于类球形水果具有较强的适用性,在产业化应用上有很大的潜力.

For the three key quality indicators of citrus fruit diameter,coloring rate and internal sugar level,the citrus comprehensive quality non-destructive testing and grading system was designed based on the double-cone roller fruit cup transmission line with feeding part,machine vision detection module,near-infrared spectroscopy detection module and grading execution part.In the machine vision detection module,the single camera was used to capture videos of rolling citrus for obtaining large number of citrus images in different postures and performing contour extraction.The fruit diameter was calculated with the average value of the minimum external circle diameter of all single citrus frame images,and the average value of its two-dimensional yellow proportion obtained by each frame image was used as the full surface coloring rate.In the near-infrared spectrum detection module,the transmission light path was designed to collect the citrus transmittance spectrum,and the mixed attitude sugar detection model was established according to the two high-frequency attitudes of citrus during on-line detection.Comparing the modeling results under different pretreatment methods,the partial least squares(PLS)model established after applying the more effective multi-scattering correction(MSC)was selected.The on-line testing results show that the maximum absolute error of fruit diameter detection is-1.42 mm with the maximum absolute error of coloring rate detection of 0.048,and the correlation coefficient of the sugar test results is 0.817 with the root mean square error of 0.658%.The joint detection and grading method of internal and external quality determines the joint grading methods of three qualities according to the decision-making method of the discriminant tree.At the sorting speed of 5/s,the average accuracy of the comprehensive grading can reach 91.16%.The overall structure of the detection and grading system is simple with strong applicability for sphere-like fruits,which has great potential for industrial application.