Research On Infrared Weak And Small Target Detection Methods Under Complex Background Conditions

Infrared detection systems have advantages such as full-time operation,strong passive detection concealment,long detection distance,and strong anti-interference ability.Infrared small target detection is a key technology in infrared detection systems.In recent years,with the rapid development of image processing and infrared imaging technology,infrared small target detection technology has been widely used in military and civilian fields such as infrared warning,infrared guidance,and industrial testing.Due to the lack of texture and structural features,small size,and low signal-to-noise ratio of infrared small targets,they are prone to being submerged in complex and variable background clutter,and are also subject to noise interference.Therefore,infrared small target detection in complex backgrounds has always been a hot and difficult research topic in the field of target detection.In order to improve the detection performance of infrared small target detection algorithms under complex background conditions and enhance the robustness of small target detection algorithms in various scenarios,this article conducts in-depth research on infrared small target detection technology.This article starts with the principles of infrared imaging and infrared imaging systems,and analyzes the structural characteristics of infrared images from three aspects: weak small targets,background,and noise characteristics.It can be concluded that the key step in detecting infrared weak small targets in complex backgrounds is to enhance the targets and suppress background and noise.This article investigates four image preprocessing methods: median,mean,Gaussian,and morphological filtering.Afterwards,the role of each layer in convolutional neural networks is analyzed,laying a theoretical foundation for subsequent research.In response to the problem that existing infrared small target detection algorithms cannot effectively enhance the target and suppress complex background interference,this paper fully considers the differences between the target and background from both local contrast and image entropy,and proposes an infrared small target detection algorithm based on entropy weighted multi-scale local contrast.Firstly,the ratio and difference forms are combined to redefine local contrast.The improved local contrast enhancement factor can significantly enhance the target while suppressing background clutter.Secondly,local entropy is used to reflect the grayscale mutation of the target area in infrared images,and an improved local entropy operator is used to weight the multi-scale local contrast,which can further highlight weak and small targets.Finally,adaptive threshold segmentation is used to separate the target from the background and obtain the final infrared weak target.The results of comparative experiments with existing algorithms show that the proposed algorithm can effectively detect infrared small targets in different backgrounds,has better target enhancement and background suppression effects,and has higher detection rates and lower false alarm rates.In response to the shortcomings of manual feature selection in traditional detection algorithms,which leads to poor robustness,and the use of sliding windows to traverse images resulting in high computational complexity,in order to improve the difficulty in describing target features and the slow detection speed of traditional algorithms,this paper selects YOLOv5 object detection algorithm with strong feature extraction ability and fast detection speed as the foundation,and proposes an improved YOLOv5 based infrared small target detection algorithm.Firstly,an additional detection layer P2 is added to the existing three detection layers for small and weak targets,enabling the network to detect smaller targets and enhancing feature information fusion between shallow and deep layers.Secondly,adding CBAM attention mechanism to the backbone network can improve the network’s attention to small targets,suppress background clutter and noise interference,and enhance the expression ability of small target channels and spatial features.Finally,change the loss function to EIo U＿Loss considers more comprehensive factors when predicting box regression,which can make the detection results more accurate.The results of ablation experiments on the experimental dataset of infrared small targets show that all three improvement strategies are effective in improving the detection accuracy of infrared small targets.The average accuracy of the improved YOLOv5 model reaches 95.48%,which is 3.66% higher than that of the original YOLOv5 model,and the detection speed can reach 46 frames per second.Compared with existing deep learning based object detection algorithms,the results of experiments show that the proposed algorithm can balance detection accuracy and detection speed,and has better object detection performance.