基于中位数绝对偏差的异常训练样本探测方法

龚循强; 张方泽; 鲁铁定; 陈志高

doi:doi:10.3788/LOP57.231202

激光与光电子学进展, 2020, 57 (23): 231202, 网络出版: 2020-12-08

基于中位数绝对偏差的异常训练样本探测方法下载： 733次

Abnormal Training Samples Detection Method Based on Median Absolute Deviation

龚循强 ^1,2,*张方泽 ^1,2鲁铁定 ^1,2陈志高 ²

作者单位

¹ 东华理工大学放射性地质与勘探技术国防重点学科实验室, 江西南昌 330013

² 东华理工大学测绘工程学院, 江西南昌 330013

图 & 表

图 1. 验证样本

Fig. 1. Verified sample

下载图片查看原文

图 2. 不同类型的训练样本。(a)不纯样本;(b)错选样本

Fig. 2. Different types of training samples. (a) Impure sample; (b) wrong choice sample

下载图片查看原文

图 3. 不纯训练样本剔除前后的分类结果。(a)剔除前;(b)剔除后

Fig. 3. Classification results before and after removal impure training samples. (a) Before removal; (b) after removal

下载图片查看原文

图 4. 错选训练样本剔除前后的分类结果。(a)剔除前; (b)剔除后

Fig. 4. Classification results before and after ramoval of wrongly selected training samples. (a) Before removal; (b) after removal

下载图片查看原文

表 1不纯训练样本剔除前后的分类精度

Table1. Classification accuracy before and after removal impure training samples

Class	Before removal		After removal
Class	Producer/%	User/%	Producer/%	User/%
Building	99.494	64.125	96.866	94.571
Vegetation	68.037	100.000	95.658	99.807
Water	71.030	84.295	99.807	97.097
Bare ground	81.883	86.316	98.573	99.711

查看原文

表 2错选训练样本剔除前后的分类精度

Table2. Classification accuracy before and after removal of wrongly selected training samples

Class	Before removal		After removal
Class	Producer/%	User/%	Producer/%	User/%
Building	94.439	62.329	97.573	98.469
Vegetation	79.676	99.884	99.307	99.583
Water	79.788	85.508	100.000	98.674
Bare ground	67.475	98.542	98.288	98.569

查看原文

龚循强, 张方泽, 鲁铁定, 陈志高. 基于中位数绝对偏差的异常训练样本探测方法[J]. 激光与光电子学进展, 2020, 57(23): 231202. Xunqiang Gong, Fangze Zhang, Tieding Lu, Zhigao Chen. Abnormal Training Samples Detection Method Based on Median Absolute Deviation[J]. Laser & Optoelectronics Progress, 2020, 57(23): 231202.

基于中位数绝对偏差的异常训练样本探测方法下载： 733次

图 1. 验证样本

Fig. 1. Verified sample

图 2. 不同类型的训练样本。(a)不纯样本;(b)错选样本

Fig. 2. Different types of training samples. (a) Impure sample; (b) wrong choice sample

图 3. 不纯训练样本剔除前后的分类结果。(a)剔除前;(b)剔除后

Fig. 3. Classification results before and after removal impure training samples. (a) Before removal; (b) after removal

图 4. 错选训练样本剔除前后的分类结果。(a)剔除前; (b)剔除后

Fig. 4. Classification results before and after ramoval of wrongly selected training samples. (a) Before removal; (b) after removal

表 1不纯训练样本剔除前后的分类精度

Table1. Classification accuracy before and after removal impure training samples

表 2错选训练样本剔除前后的分类精度

Table2. Classification accuracy before and after removal of wrongly selected training samples

关于本站 Cookie 的使用提示

全站搜索

基于中位数绝对偏差的异常训练样本探测方法 下载： 733次

图 1. 验证样本

Fig. 1. Verified sample

图 2. 不同类型的训练样本。(a)不纯样本;(b)错选样本

Fig. 2. Different types of training samples. (a) Impure sample; (b) wrong choice sample

图 3. 不纯训练样本剔除前后的分类结果。(a)剔除前;(b)剔除后

Fig. 3. Classification results before and after removal impure training samples. (a) Before removal; (b) after removal

图 4. 错选训练样本剔除前后的分类结果。(a)剔除前; (b)剔除后

Fig. 4. Classification results before and after ramoval of wrongly selected training samples. (a) Before removal; (b) after removal

表 1不纯训练样本剔除前后的分类精度

Table1. Classification accuracy before and after removal impure training samples

表 2错选训练样本剔除前后的分类精度

Table2. Classification accuracy before and after removal of wrongly selected training samples

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

基于中位数绝对偏差的异常训练样本探测方法下载： 733次