Research On Infrared Target Detection Algorithm Based On Two Bands

Target detection in infrared image is a hot research topic in infrared imaging and image processing.The core of target detection is to extract the target area of interest from the image.Infrared image target detection is widely used in military and civil fields such as security,surveillance and so on.At present,there are many studies on target detection of single-band infrared images,but the imaging effect of a single band is limited by the imaging mechanism,and is also vulnerable to background,noise and other factors,resulting in poor detection results.Therefore,how to detect the dual-band infrared fusion has great theoretical and practical application value,and has been concerned by many researchers.In this paper,the imaging characteristics of target and background in both infrared mid-wave and long-wave bands are analyzed.By analyzing the radiation characteristics of different temperature targets in two bands,it is concluded that infrared mid-wave and long-wave imaging are beneficial to the detection of high and low temperature targets,respectively.On this basis,the imaging characteristics of small target and surface target in Infrared Dual-band are studied and analyzed,and the mean gray difference and gradient direction characteristics of small target are suitable for detection.The area target has the characteristics of less detail information.The analysis results provide a theoretical basis for further research and improvement of target detection algorithm in this paper.To solve the low contrast of small infrared target,small image size and background interference resulting in low detection rate and high false alarm rate,this paper presents a dual-band detection fusion algorithm for small target,which combines improved multi-scale guided filter with directional gradient algorithm.First,based on the discontinuity between the target and background,the multiscale guided filter algorithm is improved by using local differences as weights to enhance the ability of the algorithm to distinguish the target from the background.Then,according to the feature of small target gradient isotropy,the direction gradient algorithm is used to detect the possible location of the target to reduce false alarm and complete the detection of small target in a single band.Finally,based on the detection results of each medium and long wave,a decision-level fusion algorithm is used to final confirm the target,eliminating false alarms that are not easy to remove in a single band.The simulation results show that the algorithm improves both the detection rate and the false alarm rate.To solve the problem of low detail information and low detection success rate of infrared surface target detection,a dual-band infrared surface target detection fusion algorithm is proposed in this paper.First,the YOLOv3 algorithm is trained using the marked training set of mid-wave and long-wave infrared images,respectively,to get two YOLOv3 detection networks of mid-wave and long-wave.Then,the trained network is used to detect the polygon target from the images of the validation set.Because of the different target temperature,the two detection networks have their own advantages in the detection results of different kinds of targets.Therefore,an improved non-maximum suppression algorithm is used to fuse the detection results.The experimental results show that the algorithm can overcome the problem of insufficient detail information of infrared surface target by the result of dual-band image detection,and has a higher average detection rate of face target.