基于动态网格和分区域聚类的玉米苗带识别算法研究

Research on corn seedling belts recognition algorithm based on dynamic grid and sub-region clustering

针对基于计算机视觉的玉米苗带中心线提取受自然环境干扰严重的问题,提出基于6×6动态网格与分区域特征点聚类的玉米行定位算法。首先将获取的玉米苗带图像进行像素归一化,采用改进的过绿特征和最大类间方差法分割玉米苗带与土壤背景,得到二值图像;然后通过动态网格扫描二值图像,获取候选玉米苗带特征点,并对候选玉米苗带特征点采用分区域聚类算法,得到玉米苗带特征点;最后通过最小二乘法对特征点进行线性拟合得到玉米苗带中心识别线。田间试验表明,该算法具有较好的抗干扰性能,能够很好的适应较为复杂的田间环境。玉米苗带识别准确率为93.4%,处理一幅分辨率为1920像素×1024像素的图像平均耗时320 ms。

A location algorithm of corn rows based on 6×6 dynamic grid and clustering sub-regional feature points was presented aiming at solving the problem that the centerlines extraction of corn seedling belts based on computer vision is severely interfered by natural environment.First,pixel normalize the acquired image of corn seedling belts,adopting the improved excess green algorithm and the method of maximum between-cluster variance in the segmentation of the corn seedling belts and the soil background to get the binary image;then scan the binary image with the dynamic grid,obtain the candidate feature points of the corn seedling belts,and apply the clustering sub-regional algorithm to the candidate feature points of the corn seedling belts to get the feature points of the corn seedling belts;finally using the least square method to the feature points for a linear fitting to get the center recognition line of the corn seedling belts.Field experiment shows that this algorithm enjoys a good anti-interference ability,which adapts well to relatively complicated field environment.The recognition accuracy of corn seedling belts is 93.4%,with an average time cost of 320 ms in processing an image with a resolution of 1920 pixel×1024 pixel.