基于X射线图像的干桂圆果肉质量预测

Prediction of dry longan flesh weight based on X-ray imaging

利用X射线成像技术预测干桂圆果肉质量。依次采用固定阈值法和自适应阈值法对去噪和增强后的桂圆X图像分割,分别获取桂圆区域及其核肉区域;以核肉区域质心为原点扫描该区域,检测果核与果肉之间的分界线,实现二者分割。采用基于区域形状和基于区域灰度特征的2种方法预测干桂圆果肉质量。基于区域形状预测方法是以果肉区域平均厚度、果核长短轴为基础对果核和核肉进行椭球拟合,计算二者体积差值预测果肉质量,得到训练集和预测集的相关系数分别为0.7773和0.7505,预测平均绝对偏差0.3386 g;基于区域灰度预测方法是计算图像取反后果肉区域的灰度值总和,进而预测果肉质量,得到训练集和预测集的相关系数分别为0.8304和0.7945,预测平均绝对偏差0.2929 g。分析认为基于区域灰度特征方法更能准确预测干桂圆果肉质量。

The X-ray imaging technology was applied to predict the flesh weight of dry longan. After moving noise and enhancement pretreatments, the fixed threshold and adaptive threshold were conducted to segment the longan and the flesh-seed regions. The boundary between the seed and the flesh was screened out by the algorithm that scanned from the gravity point of the flesh-region region to the flesh edge and searched the possible points which belongs to the boundary. The flesh weight was predicted by region shape-based method and the region gray-based method respectively. The region shape-based method extracted the average thickness, the axis width and length of seed region to fit the ellipsoid, and inferred the volume of flesh to predict its weight, with the correlation coefficient of 0.7773 in calibration set and 0.7505 in prediction set, as well as the average absolute deviation of 0.3386 g in prediction set. The region gray-based method extracted the total gray value from the flesh region in the inverted image to predict the weight of flesh, with the correlation coefficient of 0.8304 in calibration set and 0.7945 in prediction set, as well as the average absolute deviation of 0.2929 g in prediction set. By comparison, it was conclude that the region gray-based method could be more accurately predict the flesh weight of Longan.