Page 123 - 《广西植物》2026年第6期
P. 123
1 0 3 8 广 西 植 物 46 卷
表 5 不同特征组合模型分类精度
Table 5 Classification accuracy of models based on different feature combinations
%
特征组合 SVM 分类精度 RF 分类精度
Feature combination SVM classification accuracy RF classification accuracy
原始光谱+三边参数 70.56±7.60 56.53±8.98
Original spectrum + three ̄edge parameters (S1+S5)
原始光谱+植被指数 67.08±7.86 56.53±7.88
Original spectrum + vegetation index (S1+S6)
原始光谱+三边参数+植被指数 66.25±9.09 57.22±9.97
Original spectrum + three ̄edge parameters + vegetation index (S1+S5+S6)
原始光谱+倒数取对数 61.39±7.45 59.58±11.29
Original spectrum + reciprocal taking logarithm (S1+S2)
原始光谱+倒数取对数+三边参数+植被指数 61.39±6.88 57.36±9.60
Original spectrum + reciprocal taking logarithm + three ̄edge parameters +
vegetation index (S1+S2+S5+S6)
原始光谱+二阶导数 52.64±7.41 54.72±9.61
Original spectrum + the second derivative (S1+S4)
全特征组合 52.22±6.70 55.28±9.44
Full feature combination (S1-S6)
原始光谱+倒数取对数+一阶导数 51.53±7.09 57.92±8.63
Original spectrum + reciprocal taking logarithm + the first derivative (S1+S2+S3)
原始光谱+倒数取对数+一阶导数+二阶导数 51.11±6.63 56.11±7.96
Original spectrum + reciprocal taking logarithm + the first derivative + the second
derivative (S1+S2+S3+S4)
原始光谱+一阶导数+三边参数+植被指数 50.69±8.75 54.72±10.25
Original spectrum + the first derivative + three ̄edge parameters + vegetation index
(S1+S3+S5+S6)
仅原始光谱 48.61±6.77 54.58±10.17
Only original spectrum (S1)
原始光谱+一阶导数 48.61±8.43 55.56±9.46
Original spectrum + the first derivative (S1+S3)
倒数取对数+二阶导数 38.33±7.17 35.83±7.02
Reciprocal taking logarithm + the second derivative (S2+S4)
仅三边参数 36.11±9.76 31.25±8.72
Only three ̄edge parameters (S5)
倒数取对数+一阶导数 33.61±8.12 37.64±6.93
Reciprocal taking logarithm + the first derivative (S2+S3)
仅倒数取对数 33.33±7.30 36.81±7.22
Only reciprocal taking logarithm (S2)
一阶导数+二阶导数 26.11±5.37 30.14±6.52
The first derivative + the second derivative (S3+S4)
仅二阶导数 26.11±7.21 27.92±7.69
Only the second derivative (S4)
仅一阶导数 24.44±6.61 26.53±7.57
Only the first derivative (S3)
仅植被指数 23.19±6.52 25.00±7.22
Only vegetation index (S6)
合的分类性能呈现明显差异ꎮ 算法整体性能差异 特征组合的平均精度范围为 23.19% ~ 70.56%ꎬ均值
方面ꎬSVM 分类能力整体优于 RFꎮ SVM 算法 20 组 为 48. 95%ꎻ RF 算 法 平 均 精 度 范 围 为 25.00% ~

