| Infrared detectors are not affected by the time and the weather.They are widely used in space detection,strategic early warning,and other fields.High efficiency,robust,and reliable infrared weak target detection is a critical technology,which can improve China’s infrared early warning,anti-missile,and air control capability in a future war,and has high military value.Due to the long imaging distance of infrared images,they infrared image has low resolution and lacks detail and texture information.Then the background of the target is complex,and the target is easy to be submerged in the background,so research on infrared target recognition and detection technology is needed.The distance between the target and the infrared detector is relatively long.And the target occupies only a portion of the imaging area.It is difficult to extract significant features such as the target’s shape,structure,and texture.The useful signal is weak.Isolated noise are similar to point targets.A large number of noise points have a wide range of gray values,ranging from lower than the image’s average gray level to higher than the target.Due to high noise interference,target detection is usually under low signal-to-clutter ratio(SCR)conditions.The infrared image has a large pixel size,low image resolution,unclear targets,and background edges.And all objects above absolute zero in the spatial environment can radiate infrared waves.Therefore,it is easy to cause high-detection false alarms(FA)under complex backgrounds with clouds,atmosphere,and sunlight interference.Then,the detected target must be in a specific background.The background of the image frame must be continuous and relevant,and it is also in a dynamic change process.The image background has complexity and non-stationary characteristics,which is hard to make a proper model.During the tracking process,the target will be obscured by the background of the environment.Finally,the characteristics of small targets vary greatly.Each type of small target needs to construct corresponding datasets to reflect its essential characteristics.Otherwise,it is difficult to complete end-to-end detection efficiently.Our research focuses on the following four contents based on many algorithms in small IR target detection.First,a fusion multi-scale local gradient contrast enhancement algorithm is proposed for the detection of real infrared small targets in complex backgrounds such as large areas of high-brightness clouds and ground buildings.Combining spatial domain template filtering with multi-scale gradient contrast methods,we perform the fusion method under complex environmental backgrounds.We can suppress high-brightness clutter backgrounds in real complex airspace and improve the SCR of the target.It has the potential to improve detection accuracy.At the same time,the computational complexity is reduced.It is suitable for engineering applications.Second,a method for detecting small infrared targets in complex backgrounds is proposed,which uses the characteristics of specific infrared images with bright stripes and corners.Through a fused improved multi-scale local contrast algorithm,spatial domain filtering and an improved multi-scale local contrast method are fused multiple times at the pixel level.It could further enhance the local contrast of the target and achieve an effective background suppression factor(BSF)under different complex environmental backgrounds.It improves the signal-to-clutter ratio gain(SCRG)of the target.It has a good suppression effect on bright background edges and corners.It is strong enough to adapt to a variety of environments.Third,we apply the low rank and sparse representation to the small IR target detection and achieve a higher detection rate.The low-rank sparse representation of small targets and backgrounds in infrared images achieves a high detection rate of small infrared targets under high-noise environments.Based on the model of sparse representation of infrared block images,an improved weighted nuclear norm(WNNM)optimization algorithm is proposed to complete sparse low-rank decomposition.We adopt image local contrast as the post-detection processing.It is used for single-frame small target detection in actual infrared images.Compared to traditional spatial filtering detection and sparse representation model detection,it has better robustness against noise backgrounds.It can better suppress backgrounds and improve SCRG.Last but not the least,a low-rank sparse noise reduction algorithm based on adaptive group sparsity is implemented for processing small target images in real complex infrared scenes.Considering the characteristics of infrared images such as low SCR and low resolution,the adaptive weighted nuclear norm minimization(WNNM)group sparse model is proposed to further improve the noise reduction performance in infrared images.It has achieved noise reduction optimization from objective metrics.At the same time,the denoising performance is further verified by judging the local SCR changes of small targets in the image.Compared with the traditional denoising algorithm,it improves the peak signal-to-noise rate(PSNR)and increases the structure similarity(SSIM)in the IR image.It increases the SCR of the small target to some extent,which is useful for us to detect the interrupted object in the following target detection. |
