基于最小二乘法的早期玉米作物行检测研究

Detection of Early Crop Rows Base on Least Square Method

根据农田图像的特点,提出一种新的作物行检测算法。首先为图像设置感兴趣区域,使其仅保存两条作物行,接着用超绿法分割作物与背景,再对图像进行降噪处理。从图像每一行中选取最长的线段作为作物行,并提取中点为特征点,利用左右位置关系将特征点分为左右两类,对分类后的特征点进行线性回归,得到最终结果。通过大量的测试和与霍夫变换和随机霍夫变换的比较发现,该算法具有抗干扰强、检测速度快的优点。

A new crop row detecting algorithm is proposed acceding to features of farmland.Firstly,ROI（region of interesting）was set up to only reserve two crop rows.Then,crops and background were segmented with 2G-R-B,and then noise reduction treatment was conducted for the image.The longest line segment was chosen from each row as crop row,and midpoint was extracted as the feature point.The feature points were classified into left and right according to the position relation.Linear regression was conducted for the classified feature points.The final results show that through plentiful tests and comparison by Hough Transfom and random Hough Transfom,this algorithm has such advantages as strong anti-interference and fast detection aped.