Performance Evaluation Of Target Recognition Algorithms Based On Deep Learning

In recent years,with the deep application of deep learning to the field of automatic target recognition,how to comprehensively and objectively evaluate the performance of target recognition methods based on different depth network models is an urgent problem in practical applications.With the development and application of automatic target recognition technology,existing deep learning-based target recognition performance evaluation indicators cannot meet the current practical application requirements of target recognition technology.Therefore,this paper focuses on the problem of automatic target recognition performance evaluation.Based on the ship target recognition application,we conduct research on the key technologies related to the performance evaluation of target recognition algorithm based on deep learning.The main work is as follows:(1)Aiming at the practical application of automatic target recognition,a new automatic target recognition performance evaluation index system is designed and constructed from three dimensions of accuracy,real-time and complexity.(2)A automatic target recognition performance evaluation method based on DSER is proposed.Firstly,we use AHP method to determine the weight of each indicator in the performance evaluation index based on the questionnaire of domain experts;Secondly,the ER method is combined with the DS method and the total confidence of the algorithm performance is calculated by combining the multiple performance evaluation indexes with reliability matrix and the evidence theory features;Finally,the final performance evaluation result is calculated by the weighted method.This method has the characteristics of low computational complexity and it can quantitatively describe the uncertainty of the performance of automatic target recognition algorithm.Moreover,the method can rank the performance of multiple automatic target recognition algorithms and finally determine the automatic target recognition method with the best performance.(3)Taking the application of ship target recognition as the background,we use the existing public dataset and self-built ship dataset to carry out the example verification and analysis of six deep network automatic target recognition algorithms: in the resource limited application scenario,the performance of the improved SSD method is the best;in the resource unlimited application scenario,the yolov3 method is the best;the experimental results are consistent with the actual application.(4)Using the idea of front-end separation and integrating element UI component library,vue.js and spring MVC technology,we design and implement a prototype software for performance evaluation of target recognition algorithm based on deep learning and the experimental verification and analysis are carried out with this prototype software.