摘要
为了提升储运发箱在运输过程中行为识别的精度,提出了一种改进的特征选择方法用于降低储运发箱原始特征集的冗余程度,从而提升分类的准确率。基于滑动窗口和特征提取技术从储运发箱三轴运动数据中提取时域和频域特征;再利用改进的随机森林袋外误差方法衡量特征的重要性,并通过递归特征消除算法提取贡献度大的特征组成最优特征子集;将获得的最优特征数据集输入至分类器中,完成对储运发箱行为姿态的识别。实验结果表明,该改进的特征选择方法能够根据特征的重要性筛选出低维度且相关性高的特征子集,可以在降低分类器复杂程度的基础上提高分类的准确率,具有一定的实际工程应用价值。
In order to enhance the precision of the behavioral recognition of storage and transportation containers during the conveying process,an enhanced feature selection methodology was proposed with the objective of reducing the redundancy degree of the original feature set of storage and transportation containers,thereby improving the classification accuracy.Based on the sliding window and feature extraction techniques,time domain and frequency domain features were extracted from the three-axis motion data of the storage and transportation container.The importance of the features was then measured by the improved random forest out-of-bag error method,and the features with large contributions were extracted to form an optimal subset of the features by the recursive feature elimination algorithm.The optimal feature dataset was then inputted into a classifier,and the behavioral gestures of the storage and transportation container were recognized.The experimental results demonstrate that the improved feature selection method is capable of filtering out a feature subset with low dimensionality and high relevance in accordance with the importance of the features.This approach can enhance the accuracy of classification while reducing the complexity of the classifier,offering a valuable practical engineering application.
作者
张伟
邓士杰
于贵波
ZHANG Wei;DENG Shijie;YU Guibo(Shijiazhuang Campus of Army Engineering University,Shijiazhuang 050003,Hebei,China)
出处
《火炮发射与控制学报》
北大核心
2025年第3期32-40,共9页
Journal of Gun Launch & Control
关键词
储运发箱
特征选择
随机森林
袋外误差
递归特征消除
分类器
storage and transportation container
feature selection
random forest
out-of-bag error
recursive feature elimination
classifiers
