基于无人机可见光影像的免耕种植夏玉米苗数估算

Estimation of No-Tillage Planting Summer Maize SeedlingNumber Based on UAV Visible Light Images

为了快速估算免耕种植夏玉米出苗数,提高大田夏玉米种植管理的精准性,本研究利用无人机搭载可见光相机获取夏玉米田块高分辨率可见光影像,计算8种植被指数并结合最大类间方差法分割植被与非植被,经分析,选择红色植被指数(RI)二值化图像对可见光影像掩膜;然后统计夏玉米和杂草的24项纹理特征,比较杂草特征的变异系数及其与夏玉米的相对差异系数,选择红色方差提取夏玉米苗的特征,使用时序交点阈值法确定的阈值去除杂草干扰;提取夏玉米苗形态学特征参数作为样本,采用支持向量机(SVM)、BP神经网络、K近邻和决策树4种算法构建夏玉米苗数预测模型。结果表明,SVM和决策树算法的整体效果较好,决定系数均超过0.8且平均绝对误差(MAE)小于0.3,尤以决策树模型的精度最高,可达94.1%。本研究结果可为大面积夏玉米出苗率估测提供技术支持。

This study aimed to quickly estimate the emergence number of no-tillage planting summermaize and improve the field management accuracy.Using the high-resolution visible light images of summermaize fields obtained by UAV equipped with a visible light camera,8 vegetation indices were calculated andused to segment vegetation and non-vegetation combined with the maximum between-class variance method.After comparing the segment effect,the red vegetation index(RI)binary image was selected for the visiblelight image mask.Through counting of summer corn and weeds,comparing the variation coefficients of 24 texture features of weeds and their relative difference coefficient with those of summer maize,the red variance wasselected as the feature to extract summer maize seedlings,and the threshold determined by time series intersection threshold method was used to remove weed.With the extracted morphological characteristic parametersof summer maize seedlings,four algorithms of support vector machine(SVM),BP neural network,K-nearestneighbor and decision tree were used to construct the prediction model of summer maize seedling number.Theresults showed that the overall effects of SVM and decision tree algorithm were better with determination coefficient over 0.8 and mean absolute error less than 0.3,and the accuracy of decision tree model was the highest up to 94.1%.The results of this study could provide technical support for estimating the emergence rate ofsummer maize in large area.