基于图像滤波技术的水稻谷粒数字化测量方法

Digital Measurement Method of Rice Grain Based on Image Filtering Technology

本研究利用基于有界平均振荡模型(boundary mean oscillation, BMO)的ZQ梯度算子,结合各向异性非线性偏微分方程模型,构造用于图像处理的BMO滤波器,对水稻谷粒的数字图像进行去噪增强、边缘检测和特征提取,并对粒形参数进行统计分析。在此基础上,比较了BMO滤波与中值滤波的处理效果,并分析了BMO滤波技术的准确性与稳定性。结果表明,BMO滤波在保留图像的边界与细节特征方面显著优于中值滤波,其处理图像获取的谷粒粒长、粒宽和长宽比与人工测量值无显著差异(p <0.05),平均粒面积与千粒重正相关性强(R2 = 0.942, p <0.001),且粒形参数提取结果在不同水稻品种间有较好的稳定性。

In this paper, the ZQ gradient operator based on bounded mean oscillation (BMO) and the anisotropic nonlinear partial differential equation model were used to construct a BMO filter for image processing, which was used to denoise, detect edges and extract shape features of the digital images of rice grain. The rice grain shape parameters were statistically analyzed to compare the processing effects of the image filters and evaluate the accuracy and stability of BMO filter. The results showed that the BMO filter was superior to the median filter in retaining the boundary and detail features of the digital image. The grain length, grain width and length-width ratio obtained from images processed by the BMO filter were not significantly different from the manual measurements and there showed a strong positive correlation between the average grain area and thousand grain weight. The rice grain shape parameters obtained from images processed by the BMO filter were stable among different rice varieties. Due to the limitation of the two-dimensionality of rice grain digital images and the related equipments, it is impossible to reflect the grain shape characteristics, such as grain thickness, length-thickness ratio and so on. In future research work, the combination of image filtering technology and cross-sectional scanning technology can be considered to describe the shape characteristics of rice grain in three-dimensional space and provide accurate data for rice breeding.