Research And Application Of Infrared Object Detection Based On Deep Learning

Infrared image object detection is a basic task in the fields of infrared investigation,intelligent security,and night-time assisted driving,which provides the most basic safety guarantee for all aspects of life.The traditional infrared target detection algorithm relies on the manual design of feature extraction method,the detection accuracy is low and the scope of application is limited.However,the rapid development of deep learning technology provides a new research idea and direction for infrared image object detection,which can extract robust semantic features and greatly improve the overall detection level.Therefore,the research on the improvement and optimization of the object detection method based on deep learning for infrared images is of great significance to improve the accuracy of object detection and model generalization.The main research work of this thesis is as follows:(1)In order to solve the problems of low signal-to-noise ratio and insufficient data set of infrared images,the image denoising and enhancement algorithms are studied.Firstly,the infrared images is preprocessed by CBDNet,a denoising algorithm based on deep learning.Secondly,the existing data sets are expanded by basic image transformation.Finally,the mosaic data enhancement algorithm is used to improve the target and background richness of the data sets.(2)In this thesis,an infrared image object detection algorithm based on attention mechanism and multi-scale feature fusion is proposed to solve the problems of lack of detail information and non-obvious target features in infrared images.The algorithm achieves efficient feature extraction by adding convolutional attention model CBAM to the backbone feature extraction network and adopting an improved 3-point Bi FPN multi-scale feature fusion method,and further utilizes the improved NMS optimization algorithm design.(3)On the basis of the above research,a object detection algorithm based on the decision-level fusion of infrared and visible images is proposed,and the fusion of infrared and visible images can achieve information complementarity.This algorithm combines the advantages of infrared images that are not susceptible to light and the advantages of rich detailed information of visible image targets,which further improves the detection accuracy of the algorithm.At the same time,for the problem of high computational complexity of the algorithm,the algorithm parameters are optimized by applying deep separable convolution on the basis of(2).(4)In order to meet the practical application needs of infrared object detection,an easy-to-use human-computer interaction interface is built using the Py Qt framework,and an infrared video surveillance platform was designed.The experiment in this thesis shows that the improved infrared image object detection algorithm can improve the detection accuracy by 2.62%,the parameter amount can be reduced by 42.8% after the application of deep separable convolution,and the object detection algorithm that integrates infrared and visible images at the decision level is further improved in detection accuracy.