复杂条件下多玉米籽粒识别与统计方法研究

Study on Identification and Statistics of Corn Seeds under Complex Condition

为提高复杂条件下多玉米籽粒的识别与统计效果,以图像处理为手段,在非均匀光照校正和噪声滤除的基础上,提出了新的图像阈值搜索范围和新的阈值判别准则,以寻找最佳阈值。采用改进的最大类间方差(Otsu)算法对玉米籽粒图像进行分割,结合基于连通区域的面积法对籽粒数目进行统计。试验结果表明:研究的新算法不仅缩短了运算时间,并且有效的去除了非均匀光照和噪声等因素的干扰,提高了玉米籽粒识别与统计的准确率。通过对50幅玉米籽粒图像进行算法测试,得出籽粒识别的平均准确率达到97%,说明了研究提供的籽粒分割与统计方法准确率高,对于种子千粒重等物料特性分析有重要的应用价值。

In order to increase effects of identification and numerical statistics of corn seeds under complex condition, non-uniform illumination was corrected and noise was filtered based on image treatment technology, and a new threshold searching range and a new threshold criterion for finding optimal threshold were developed. Maximum between-cluster variance(Otsu)algorithm was used for image segmenting, the number of corn seeds was calculated using area method based on connected regions. The experiment results showed that the new algorithm not only shortened operation time, but also reduced the interference of non-uniform illumination and noise effectively, so as accuracy of corn seeds identification and number statistics were increased. Experiment with 50 corn seeds images verified that average seeds number accounting accuracy was up to 97%,which illustrates the research provided the method of grain split with high accuracy and had important application value for the material properties of seed, such as 1000-kernal weight.