基于机器视觉的散装粮随机扦样方法研究  

Research on Random Sampling Method of Bulk Grain Based on Machine Vision

作　　者：李智 但乃禹[1,2] 李磊 杨卫东 陈卫东[1,2] Li Zhi;Dan Naiyu;Li Lei;Yang Weidong;Chen Weidong(Henan Key Laboratory of Grain Photoelectric Detection and Control,Zhengzhou 450001;College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001)

机构地区：[1]河南省粮食光电探测与控制重点实验室,郑州450001 [2]河南工业大学信息科学与工程学院,郑州450001

出　　处：《中国粮油学报》2023年第2期130-137,共8页Journal of the Chinese Cereals and Oils Association

基　　金：河南省杰出青年基金项目(222300420004);河南省重大公益专项(201300210100);国家重点研发计划项目(2017YFD0401001-02)。

摘　　要：针对粮食收购过程中扦样设区选点不科学、不合理导致的扦取样品代表性不足和存在人为舞弊风险的问题,提出一种结合双目视觉与图像分割技术的散装粮随机扦样方法。首先使用双目相机获取装粮区域图像信息并校正,利用Unet网络模型实现校正后左图像目标分割,再使用Opencv计算目标区域4个角点像素坐标,根据扦样规则将目标区域划分为多个扦样区域并随机生成扦样点,最后针对BM匹配算法生成视差图效果较差的问题,采用SGBM(Semi-Global Matching)半全局立体匹配算法对校正后左右图像立体匹配,根据匹配结果完成扦样点三维空间定位。实验结果表明,所述方法针对装粮区域有较好的识别效果,并且实现了在3 m范围内扦样点的随机选取与定位,对粮食扦样环节的自动化和智能化发展提供了技术支撑。To solve the problem of inadequate representation of samples and the risk of human fraud caused by unscientific and unreasonable selection of sample points in grain purchasing process, a random sampling method for bulk grain based on binocular vision and image segmentation technology was proposed. Firstly, the binocular vision a was used to acquire the image information of grain loading area and correct it, and the Unet network model was used to achieve target segmentation of the corrected left image. Then, the Opencv was used to calculate the pixel coordinates of four corner points of the target area. According to the sampling rules, the target area was divided into multiple sampling areas and several sampling points were randomly generated. Finally, aiming at the problem that the poor effect of the disparity map generated by BM matching algorithm, the semi-global stereo Matching algorithm of SGBM(Semi-global Matching) was used to conduct the stereo-match of the corrected left and right images, and the 3D space positioning of the sample points was completed based on the matching results. The experimental results have shown that the method has a good identification effect for the grain loading area, and random selection and positioning of sampling points within 3 m were achieved, which provides a technical support for the automation and intelligent development of grain sampling.

关 键 词：机器视觉 图像分割 双目视觉 立体匹配 空间定位 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]