Analysis Of Evaluation Indexes And Algorithm Design Of Visible Light And Infrared Image Fusion For Target Detectio

With the development of visible and infrared image sensors,visible and infrared images have been widely used in the fields of target detection and remote sensing interpretation.Due to the influence of environment and other factors,a single visible or infrared image can not well meet the application requirements.Therefore,how to effectively mine the target features in visible and infrared images and integrate them has become a hot and difficult problem in current research.The existing visible and infrared image fusion algorithms usually only evaluate the pros and cons of the algorithms based on the quantitative index of the fusion image,ignoring the practical application value of the algorithms themselves.In order to solve the problem of weak correlation between fusion evaluation index,fusion algorithm and application,this paper takes object detection as an example to carry out a series of studies.The main work is as follows:(1)At present,visible and infrared image fusion evaluation system only evaluates the fusion image itself,resulting in insufficient relevance between the fusion evaluation system and target detection.To solve this problem,a correlation analysis method of visible and infrared image fusion evaluation index based on target detection is designed in this paper.Firstly,visible and infrared image data sets are fused by different algorithms,and the fusion results are obtained,and the qualitative and quantitative evaluation are carried out.Secondly,the main target detection algorithm is selected to detect the image after fusion,and the general detection result evaluation index is used for quantitative evaluation,and the performance of each fusion algorithm applied in target detection is analyzed.Finally,the correlation analysis between the quantitative evaluation of image fusion and the quantitative evaluation of target detection based on fusion is carried out to analyze the strength of the correlation between each fusion index and the target detection results.In this paper,visible light and infrared image data sets were made,15 representative fusion algorithms and 13 evaluation indicators were selected for fusion evaluation,and YOLO v3 and YOLO v5 as well as average accuracy and error rate were selected for the target detection process.Spearman rank correlation coefficient was used to complete the correlation analysis.The results show that the fusion index based on information theory has low correlation with the target detection results,while the evaluation index based on visual fidelity and edge detection has high correlation with the target detection results.(2)At present,the focus of the design of visible and infrared image fusion algorithm is to meet the favorable evaluation index of the algorithm itself,rather than the subsequent application,which leads to the problem that the fusion algorithm designed for the subsequent application is not suitable for the subsequent application.In order to solve this problem,the fourth chapter of this paper designs a visible light and infrared fusion algorithm for target detection.This algorithm combines image adaptive enhancement with independence(U),focus(F),object(O)significance detection,accurately controls visible and infrared image input,enhances image features,improves fusion information,and facilitates target detection.Firstly,the adaptive enhancement algorithm is added to the visible image to improve the visibility of image texture details,and the infrared image is normalized.Secondly,the processed image is decomposed into detail layer and base layer by guided filtering,and the significance detection is used to generate the weight map of detail layer,so as to improve the accurate fusion amount of visible image background information and infrared image edge information in detail layer.Finally,the final fusion image is obtained by combining the detail layer and the base layer after fusion according to the weight value.In order to verify the superiority of the proposed algorithm,the fusion evaluation indexes which are strongly correlated with the target detection results in Chapter 3 are selected for quantitative analysis of the fusion image,such as six fusion evaluation indexes such as average gradient,edge intensity,spatial frequency and visual fidelity,and the YOLO v5 network is used for target detection of each fusion algorithm.The results show that the proposed algorithm achieves the best performance in the fusion of qualitative evaluation,quantitative evaluation and target detection evaluation index average accuracy.The experimental results show that the two design schemes proposed in this paper can solve the problem of lack of correlation between visible light and fusion algorithm and application to a certain extent,and provide a suitable fusion framework for target detection,so that visible light and infrared fusion images can obtain rich information while increasing the detectability.It can provide some ideas and reference value for the application of fusion algorithm in remote sensing interpretation,target recognition and other fields.