摘要
根据田间作物垄点像素矩阵特点,基于行向量目标像素良序子集,先进行垄点子集预处理,然后运用最近邻判别准则搜寻每个垄点像素的最近邻点。通过设置最近邻搜索方向角和最近邻阈值,对断垄和较大面积的杂草等噪声影响进行控制。实验结果表明,与传统的最近邻算法比较,该算法的准确性和鲁棒性均得到提高,时间复杂度较小,对农田视觉导航实际应用有一定价值。
The conventional nearest neighbor（CNN） classifiers provide a simple approach with good robustness, which is guaranteed to converge to a result, but it has some shortcomings such as aimless searching, much time consumption, and an unexpected infection by noises and so on. In this paper, a new approach to detect crop rows was proposed, which was based on the well-ordered subsets of the objective in row vectors of image matrix. And the nearest neighbor search angle and the nearest neighbor distance which are considered important control factors were embedded in the CNN. Combined with the inherent property of crop pixels, the nearest neighbor query can be limited in a small suitable range. The experimental results indicate the algorithm was of good robustness and accuracy compared with the CNN, and it could avoid the impact of weeds with small time consumption.
出处
《中国图象图形学报》
CSCD
北大核心
2007年第11期2048-2051,共4页
Journal of Image and Graphics
基金
国家自然科学基金项目(60574029)
关键词
良序集
最近邻法
垄行识别
视觉导航
well-ordered, nearest neighbor rule, crop rows indentification, machine vision navigation
