基于机器视觉的大豆机械化收获质量在线监测方法

Online Monitoring Method of Mechanized Soybean Harvest Quality Based on Machine Vision

摘要:

针对大豆机械化收获过程中缺少联合收获机作业质量（破碎含杂率）在线监测装置的问题，提出了基于机器视觉的大豆机械化收获图像采集系统、大豆成分分类识别算法和谷物联合收获机作业质量监测方法。采用改进分水岭算法对大豆图像进行有效分割，筛选RGB和HSV颜色空间特征值，基于颜色特征值对分割后大豆图像各闭合区域进行分类识别，构建了量化评价模型，测试了算法的准确性，并进行了相关的田间试验。结果表明，R、S、H分量一阶矩特征值对大豆各成分具有较好的特征分离性，通过这3个分量颜色阈值能够很好地进行大豆成分分类；系统大豆完整籽粒查准率为87.26%、查全率为86.17%，大豆破碎籽粒查准率为86.45%、查全率为79.42%，大豆杂质查准率为85.19%、查全率为83.69%；在田间测试过程中，本文设计的检测方法对谷物联合收获机作业质量性能评定结果与人工检测一致。本文所提出的算法能快速、有效、稳定地识别完整籽粒、破碎籽粒和杂质，量化模型能准确计算出破碎含杂率，从而实现大豆机械化收获质量可视化监测与报警，可为智能谷物联合收获机参数在线监测及自适应控制策略研究提供技术支持。

Abstract:

In order to solve the problem of lack of on-line monitoring device for the quality of soybean during mechanized harvesting, the methods of image acquisition, soybean component identification and quality monitoring for mechanized harvesting were presented based on machine vision. The improved watershed algorithm was used to segment the soybean image effectively, and the color spatial characteristic values of RGB and HSV were used to classify and identify the components of the soybean image. The image acquisition system can acquire a clear soybean image in real time, segment and identify each component of the soybean sample, and calculate the real-time crushing impurity rate of mechanized harvest. The quantitative evaluation model was constructed, and the accuracy of the algorithm was tested and field experiments were carried out. The results showed that the accuracy of whole soybean seeds was 87.26% and 86.17%, respectively. The accuracy of crushed soybean grains was 86.45%, and the recall rate was 79.42%. The detection rate of soybean impurities was 85.19% and 83.69% respectively. The results of quality and performance evaluation of grain combine harvester were consistent with that of manual inspection. The results showed that the proposed algorithm can quickly, efficiently and stably identify intact grains, broken grains and impurities, and the quantization model can accurately calculate the fraction of broken impurities. At the same time, the soybean image acquisition system designed can replace the manual detection, and became an effective method for the quality evaluation of soybean combine harvester. Additionally, it can provide real-time data of the crushing miscellaneous rate of soybean during mechanized harvesting, realize visual monitoring and alarm, and provide data support for parameter adjustment of intelligent combine harvester, so as to improve the quality of soybean mechanized harvesting.