基于机器学习的迷彩伪装效果评价方法
Camouflage effect evaluation method based on machine learning
王晨 1牛春晖 1杜向坤 1刘鑫1
作者信息
- 1. 北京信息科技大学仪器科学与光电工程学院,北京 100192
- 折叠
摘要
针对迷彩伪装效果评价算法中评价指标权重分配复杂性,算法开发平台灵活性的问题,提出了基于多特征指标决策树的评价方法.该方法依据视觉注意力机制选择纹理、颜色、亮度、结构相似度与伪装目标尺寸这5项特征作为评价指标,使用机器学习决策树分类器训练出迷彩伪装效果评价模型,将模型移植入体积小、功耗低的树莓派开发平台上.通过与均值权重法、熵权法两种评价方法进行准确率对比实验,其中均值权重法准确率为56％;熵权法准确率为74％;该方法准确率为90％.通过实时性实验证明该方法可以在场外2 s左右得到迷彩伪装效果评价结果.
Abstract
In this paper,a new evaluation method based on a multi-feature indicator decision tree is proposed to ad-dress the complexity in weight allocation for evaluation metrics and the flexibility of algorithm development platforms in camouflage effectiveness evaluation.The method selects five features,texture,color,brightness,structural similarity,and camouflage target size,as evaluation indicators based on visual attention mechanisms and trains a camouflage ef-fectiveness evaluation model using a machine-learning decision tree classifier,which is ported to a small-sized,low-power Raspberry Pi development platform.Through the accuracy comparison experiment with two evaluation methods of mean weight method and entropy weight method,the accuracy of mean weight method is 56％,the accuracy rate of entropy weight method is 74％,and the proposed method achieves an accuracy of 90％.The real-time experiments demonstrate that the method can get the evaluation results of camouflage effect in about two seconds outside the field.
关键词
数字图像处理/多特征指标/树莓派/决策树/迷彩伪装效果评价Key words
digital image processing/multi-feature index/raspberry pie/decision tree/camouflage effect evaluation引用本文复制引用
基金项目
"十四五"预研项目(315087608)
北京市自然基金青年项目(4224094)
出版年
2024
NETL
NSTL
万方数据