转盘式板栗可见/近红外光谱检测分选系统研究

Research of Rotary Sorting System for Chestnuts Based on Visible/Near-infrared Spectroscopy

摘要:

针对典型木本粮食板栗的现场非破坏快检分选需求，基于可见/近红外光谱检测分析技术，设计了转盘式板栗在线检测分选系统。该系统主要由上料单元、光谱采集单元、控制单元、分选单元和显示单元组成，上料单元振动盘经试验确定最佳频率为150Hz，并联合伺服电机对转盘进行驱动，光谱采集单元以卤素灯为光源，应用微型光谱仪采集光谱信息，控制方案经对比后确定并行式执行方案，基于QT平台及C++语言开发单片机程序，配合上位机开发的光谱采集软件，实现了光谱信息采集、处理、显示存储以及分类，分选执行机构采用电磁推杆带动挡板伸缩完成孔洞的开合。以迁西、丹东、玉溪板栗为对象，采集光谱并对比不同预处理方法分别建立产地与霉变的偏最小二乘判别（Partial least squares discriminant analysis，PLS-DA）模型，交叉验证后产地模型预测集准确率为97.12%、霉变模型预测集判别准确率为94.74%。最后将模型植入系统软件后应用随机预测集板栗进行测试，产地与霉变判别分选准确率分别为93.83%和94.12%，最优检测分选速度为37颗/min。结果表明，本文设计的转盘式板栗检测分选系统效率高、采集稳定、检测准确率高，具备快速实现不同产地以及霉变板栗无损检测分选功能。

Abstract:

There is a certain demand to carry out the on-site rapid and non-destructive detection and sorting of typical woody grain, i.e. chestnuts. Based on visible/near-infrared spectroscopy and analysis technology, a rotary chestnut online detection and sorting system was developed. This system mainly consisted of a feeding unit, a spectral acquisition unit, a control unit, a sorting unit, and a display unit. The experiment of vibration disk in the feeding unit determined that 150Hz was the optimal frequency, and a servo motor was connected to drive the disk. The halogen lamps were used as light sources and a micro-spectrometer was used to collect spectral information in the spectral acquisition unit. After comparison, a parallel control plan was determined in the control unit. Based on the QT platform and C++ language, a microcontroller program was developed. The program combined with the developed spectral acquisition software for the upper computer was used to achieve the collection, processing, display, storage, and discrimination of spectral information. The electromagnetic push rod was applied to drive the expansion and contraction of the baffle to complete the opening and closing of the hole. Chestnuts from Qianxi, Dandong, and Yuxi were selected as experimental samples, spectra were collected and spectral preprocessing methods were applied to establish partial least squares discriminant analysis (PLS-DA) models for comparison. The PLS-DA models for geographical origins and mildew discrimination were developed, respectively. The correct classification accuracy of the model for geographical origins in prediction set was 97.12%, and it was 94.74% of the model for mildew discrimination. The optimal detection and sorting efficiency was to process 37 samples per minute by this device. Finally, the models were individually implanted into the system software, and random prediction sets of chestnuts were used for tests. The correct classification accuracies of the models for geographical origins and mildew discrimination achieved 93.83% and 94.12%, respectively. The results indicated that the designed rotary detection and sorting system and device presented high efficiency, stable collection, and high detection accuracy. It was feasible to quickly achieve the goal of non-destructive detection and sorting of chestnuts from different geographical origins and mildew or not.