Journal of Computer Applications

    Next Articles

Object detection uncertainty measurement scheme based on guide to the expression of uncertainty in measurement

  

  • Received:2024-07-08 Revised:2024-10-03 Online:2024-11-19 Published:2024-11-19

基于测量不确定度表示指南的目标检测不确定度测量方案

蒋沛宇1,王永光2,任亚亭1,李硕晨1,谭火彬1   

  1. 1. 北京航空航天大学
    2. 北京航天计量测试技术研究所
  • 通讯作者: 蒋沛宇

Abstract: Aiming at the problem of treating uncertainty modeling as a step for optimizing prediction results while ignoring the value of uncertainty, a scheme based on Guide to the Expression of Uncertainty in Measurement (GUM) was proposed. Firstly, sources of uncertainty in object detection algorithms were decomposed into three mutually independent aspects: data, model and platform. Uncertainty factors were extracted from these three aspects to construct measurement function. Secondly, Type A evaluation and Type B evaluation were used to quantify uncertainty influence components. Finally, uncertainty synthesis rules were used to synthesize standard uncertainty from uncertainty components. Experiments conducted on an object detection algorithm results show that compared to peak signal-to-noise ratio and structural similarity, data respectively increases by 5.30 percentage points and 19.08 percentage points in capturing noisy data; model uncertainty has a minimal effect on the prediction results, which can be ignored within the range of 10^-6; platform uncertainty can better characterize prediction result differences caused by software and hardware platforms.

Key words: object detection algorithm, uncertainty measurement, uncertainty breakdown and combination, image data, software and hardware platform

摘要: 针对当前目标检测算法不确定性研究将不确定性建模作为优化预测结果的一个步骤,忽视了不确定性本身的性质的问题,提出了一种基于测量不确定度表示指南(GUM)的方案。首先,将目标检测的不确定度来源分解为数据、模型、平台三个互相独立的方面,从三个方面提取不确定度影响因素,构建不确定度测量函数;其次,使用GUM法中的A类评定方法和B类评定方法,对不确定度影响分量进行度量;最后,基于测量函数使用不确定度合成法则,由不确定度分量合成标准不确定度。在目标检测算法上展开实验,结果表明,与峰值信噪比和结构相似性相比,数据的不确定度在捕捉噪声数据方面分别提高了5.30和19.08个百分点;模型不确定度对预测结果的影响很小,在10^-6范围内可以忽略;平台的不确定度可以用数值化形式表示由软硬件平台带来的预测结果差异。

关键词: 目标检测算法, 不确定度测量, 不确定度分解与合成, 图像数据, 软硬件平台

CLC Number: