基于超像素分割的田间小麦穗数统计方法

杜颖,蔡义承,谭昌伟,李振海,杨贵军,冯海宽,韩东

Field Wheat Ears Counting Based on Superpixel Segmentation Method

DU Ying,CAI YiCheng,TAN ChangWei,LI ZhenHai,YANG GuiJun,FENG HaiKuan,HAN Dong

表1 不同类型分类器特征

Table 1 Different types of classifier features

分类器类型 Classifier type 预测速度 Prediction speed 内存占用 Memory usage 解释性 Interpretability 模型灵活性 Model flexibility 线性支持向量机 linSVM 二分类：快 Binary: Fast 多分类：中 Multiclass: Medium 中等 Medium 简单 Easy 低 Low 在类之间进行简单的线性分隔 Makes a simple linear separation between classes 二次多项式支持向量机 quaSVM 二分类：快 Binary: Fast 多分类：慢 Multiclass: Slow 二分类：中等 Binary: Medium 多分类：大 Multiclass: Large 困难 Hard 中等 Medium 三次多项式支持向量机 cubSVM 二分类：快 Binary: Fast 多分类：慢 Multiclass: Slow 二分类：中等 Binary: Medium 多分类：大 Multiclass: Large 困难 Hard 中等 Medium 细高斯支持向量机 finGSVM 二分类：快 Binary: Fast 多分类：慢 Multiclass: Slow 二分类：中等 Binary: Medium 多分类：大 Multiclass: Large 困难 Hard 高,随内核刻度设置而减小 High, creases with kernel scale setting 类之间精细区分,内核刻度为sqrt(P)/4 Makes finely detailed distinctions between classes, with kernel scale set to sqrt(P)/4 中度高斯支持向量机 medGSVM 二分类：快 Binary: Fast 多分类：慢 Multiclass: Slow 二分类：中等 Binary: Medium 多分类：大 Multiclass: Large 困难 Hard 中等 Medium 中度区分,内核刻度为sqrt(P) Medium distinctions, with kernel scale set to sqrt(P) 粗高斯支持向量机 coaGSVM 二分类：快 Binary: Fast 多分类：慢 Multiclass: Slow 二分类：中等 Binary: Medium 多分类：大 Multiclass: Large 困难 Hard 低 Low 在类之间粗区分,内核刻度为sqrt(P)*4,其中P为预测因子数 Makes coarse distinctions between classes, with kernel scale set to sqrt(P)*4, where P is the number of predictors 细 K最近邻 finKNN 中 Medium 中等 Medium 困难 Hard 类之间细微差异区分,邻域数设为1 Finely detailed distinctions between classes. The number of neighbors is set to 1 中度 K最近邻 medKNN 中 Medium 中等 Medium 困难 Hard 类之间中等差异区分,邻域数设为10 Medium distinctions between classes. The number of neighbors is set to 10 粗 K最近邻 coaKNN 中 Medium 中等 Medium 困难 Hard 类之间粗略差异区分,邻域数设为100 Coarse distinctions between classes. The number of neighbors is set to 100 余弦 K最近邻 cosKNN 中 Medium 中等 Medium 困难 Hard 使用余弦距离度量,在类之间中等区分,邻域数设为10 Medium distinctions between classes, using a cosine distance metric. The number of neighbors is set to 10 三次多项式 K最近邻 cubKNN 慢 Slow 中等 Medium 困难 Hard 使用立方距离度量,在类之间中等区分,邻域数设为10 Medium distinctions between classes, using a cubic distance metric. The number of neighbors is set to 10 加权 K最近邻 weiKNN 中 Medium 中等 Medium 困难 Hard 使用权重距离度量,在类之间中等区分,邻域数设为10 Medium distinctions between classes, using a distance weight. The number of neighbors is set to 10