Over-sea Small Target Detection Using Infrared Images Based On Deep Learning

The detection of maritime targets is of great significance in ensuring the safety of navigation and the implementation of maritime search and rescue.It has long been widely concerned by scholars at home and abroad.Because the small target at sea is far away from the imaging system.With small area,low signal-to-noise ratio and the sea state changes are complex,the detection and tracking of small infrared target images at sea has gradually become one of the core technologies in this field.On the basis of summarizing and drawing on the research work of the predecessors,this research takes the infrared small target detection problem as the main research content,and aims to improve the detection accuracy and detection efficiency of the small infrared target at sea.Based on the study of existing research results,the corresponding improved algorithms are proposed from two aspects:infrared image preprocessing and infrared image small target recognition.The main research work and research results include:(1)The research status and representative research methods of infrared small target detection are collated,analyzed and summarized.The main problems and challenges in the detection of small infrared targets at sea are proposed.(2)In the aspect of infrared image preprocessing,the imaging principle of the infrared image and the visible image in the image,the histogram feature of the image and the image noise characteristics are analyzed.An infrared image target enhancement algorithm combining histogram equalization and Retinex is proposed.The algorithm uses wavelet transform to layer the background and target of the infrared image.The details of the infrared image are preserved in the high frequency sub-band image,the background is retained in the low-frequency sub-band image,and then the high-frequency part is enhanced by Retinex algorithm.The high-frequency partial enhancement image capable of retaining the original detail information of the infrared image is used to equalize the low-frequency partial image using the histogram equalization,thereby improving the overall contrast of the image,and then performing image weighted fusion.Experiments show that the improved algorithm can improve the image contrast and enhance the details to obtain larger information entropy,which is effective and superior in image enhancement.(3)In the small target detection of infrared image,based on the comprehensive comparative analysis of the classical small target detection algorithm,combined with the specific characteristics of the small infrared target image,a Faster R-CNN network based on improved loss function is proposed.A method for detecting small infrared targets at sea.In terms of loss function design,the improved loss function can adjust the weight of positive and negative samples,and control the weight of difficult and easy-to-sort samples,so that the model converges more quickly.The experimental results on the dataset show that compared with the performance of the original network model,the improved network based on Faster R-CNN effectively improves the detection accuracy and the convergence is faster.