基于线阵扫描图像的玉米果穗性状检测技术

Detecting Techniques of Maize Ear Characters Based on Line Scan Image

摘要:

为实现玉米果穗性状自动检测，以线阵扫描方式获取玉米果穗表面圆周图像信息，利用图像处理技术从中提取相关性状参数。从扫描图像自身提取检测模板，应用归一化互相关算法完成玉米果穗单周图像的精确切割；以Otsu法完成籽粒分割后，利用边界非完整籽粒的连接性，完成边界分割籽粒的合并，提高籽粒计数准确度；采用图像投影曲线平滑后求极值的方法获取穗行数。与人工检测实验结果对比，穗粒数和穗行数的计数准确率分别为94.6%和99.1%。

Abstract:

To acquire the properties of ear characters in maize automatically, a panorama of the maize ear was captured by using a line scan camera. Then the data was extracted through image processing methods. A part of the original line scan image was selected as a template. The ear image with exactly 360° views was cropped according to the normalized cross correlation of the template with the original image. Otsu method was used to segment the maize kernels, but the kernels on the image borders were split into two parts. The connectivity of those two parts of one kernel on both borders was taken as a clue to merge them into ones. This process improved the counting accuracy of kernel numbers. The row number of the kernel was obtained from the image projection method. The profile of projection was smoothed to solve the extreme points as the row number. Experiment results showed the counting accuracy of kernels numbers and kernel rows numbers were 94.6% and 99.1%, respectively.