工程应用中的目标识别方法评估研究

Research on Evaluation of Target Recognition Method in Engineering Application

当下目标识别方法主要通过检测经典数据集,根据准确率(如mAP)的高低或者与其他算法进行比较来评判其优劣。但在工程应用中,通常没有公开数据集参考,并且只考虑准确率等指标而不考虑其应用要求是不够的,还需要考虑多方面因素,如算力要求、处理速度、部署平台等,先前的评估方法已无法满足工程需求,因此提出了一种新的综合评估方法。首先介绍了目标识别方法的分类方法类型,然后选取基于此的两种常用的经典目标识别方法,进行简要介绍以及优缺点对比,在此基础上结合科学研究项目评估方法提出了一套可量化的目标识别方法评估标准与方法。通过设计实验验证其评估结果与经验相符,对工程应用中的目标识别方法择优有一定的参考价值。

The current target recognition method mainly judges its pros and cons by detecting classic data sets,according to the accuracy rate(such as mAP)or comparing with other algorithms.However,in engineering applications,there is usually no public data set reference,and it is not enough to only consider indicators such as accuracy without considering its application requirements.It is also necessary to consider various factors,such as computing power requirements,processing speed,deployment platform,etc.The evaluation method has been unable to meet the engineering needs,so a new comprehensive evaluation method is proposed.This paper first introduces the classification method types of target recognition methods,and then selects two commonly used classic target recognition methods based on this,briefly introduces and compares the advantages and disadvantages.On this basis,combined with scientific research project evaluation methods,a set of quantifiable Evaluation criteria and methods for target recognition methods.The design experiment verifies that the evaluation results are consistent with experience,which has certain reference value for the optimization of target recognition methods in engineering applications.