基于计算机视觉的柠檬尺寸和质量估测方法研究

Research on Lemon Size and Quality Estimation Method Based on Computer Vision

为了解决通过无接触的方式估测柠檬尺寸质量的问题,进而提高柠檬分类水平。采用相机作为采集工具获取柠檬果实和小尺寸标定板图像,利用MATLAB的Camera Calibration工具进行相机标定获取相机参数并对图像进行校正,采用双边滤波和Chroma King Technique(CK法)图像分割方法获取图像目标区域,配合计算机视觉技术估测柠檬果实长度、宽度和面积数据。在柠檬果实图像上绘制出长度六等分的网格线并获取相应数据,将网格线数据与柠檬果实尺寸数据作为参数,通过设计主实验和多个对照实验,从是否降维、不同角度图像组合、输入参数变化等多个因素进行实验比较,探索柠檬尺寸质量的估测方法。结果表明,每组实验利用两个角度柠檬果实图像,以果实长度、宽度、面积数据和网格线数据作为参数,采用两个角度图像并对特征降维后的模型进行分析,柠檬果实尺寸估测值均方根误差为8.805 cm^(3),平均绝对误差百分比为6.937%,相对误差界为7.215%,拟合优度R~2为0.757;质量估测值均方根误差为6.589 g,平均绝对误差百分比为7.974%,相对误差界为7.192%,拟合优度R~2为0.835。该研究为后续将计算机视觉技术运用于机械自动分拣技术提供理论参考依据。

In order to solve the problem of estimating the size and quality of lemons in a non-contact way,and to improve the level of lemon classification.The camera is used as the acquisition tool to obtain the images of lemon fruit and small-size calibration plate,the camera calibration tool of MATLAB is used to calibrate the camera to obtain the camera parameters and correct the images,the bilateral filtering and Chroma King Technique(CK method)image segmentation method are used to obtain the target area of the image,and the length,width and area data of lemon fruit are estimated with computer vision technology.The grid line data and the lemon fruit size data are used as parameters,and the main experiment and multiple control experiments are designed to compare the experimental results from multiple factors such as whether the dimension is reduced,the combination of images from different angles,and the change of input parameters,so as to explore the estimation method of lemon size quality.The results show that the root mean square error of the estimated lemon fruit size is 8.805 cm^(3),the mean absolute percentage error is 6.937%,the bounds of relative error is 7.215%,and the goodness of fit R_(2) is 0.757,and the root mean square error of the quality estimation is 6.589 g,and the average absolute error percentage is 7.974%,using the two-angle lemon fruit image and the grid line data as parameters.The bounds of relative error is 7.192%,and the goodness of fit R_(2) is 0.835.This study provides a theoretical reference for the subsequent application of computer vision technology to mechanical automatic sorting technology.