Research On Infrared Small Target Detection Algorithm Based On Visual Saliency

Aiming at the problems of short action distance and long reaction time of infrared surveillance and alarm systems,robust infrared small target detection technology as a key technology in this system,can not only increase its action distance,but also help to improve its reaction speed and timely find and lock the target,which is very important for the longdistance monitoring and early warning of targets.However,in some practical applications,scene in the field of view is usually complex and changeable,containing a lot of background clutter and noise,which brings a lot of interference to the detection of targets,resulting in the appearance of many false targets.While the real targets appear in infrared images in the form of small targets with dots or patches,lacking of clear texture and contour information,and have a certain scale change with the change of the detection field of view and its own movement,which makes their single-frame detection quite difficult.Especially in the environment with extremely low signal-to-noise ratio,small targets are easily annihilated in the undulating background,which further increases the difficulty of detection.In view of the above problems,this paper makes an in-depth study on infrared small target detection algorithms in complex scenes,and carries out the following work based on visual saliency:(1)A reweighted global low-rank decomposition detection model is established.Aiming at the problem that the small targets in complex scenes are difficult to detect in a single frame due to the presence of background clutter and noise interference,this paper starts from the perspective of visual saliency detection and improves the IPI model,finally establishes a reweighted infrared small target detection model based on the low-rank sparse decomposition theory of matrix.Simultaneously,the inexact Lagrange multiplier method is used for a rapid solution of this model,which effectively separates the target from the background and achieve the suppression of most background areas as well as the significant enhancement of small targets.Compared with the salience maps of other saliency detection models,the salient map obtained by this method is more stable and reliable,which shows that it is more adaptable to complex scenes.(2)Optimization of detection performance.To solve the problem that there is no significant difference of the saliency value and the low distinction between small targets and the background residual,first,a simple linear iterative clustering algorithm is used to block the image reasonably and reduce the complexity of subsequent operation.Then based on the background continuity and the local isolation of the target,a simple and effective similarity measurement strategy is used to redistribute the saliency values of each part in the salient map,which further improves the inhibition ability of the background and the enhancement of small targets,as well as achieving the complete segmentation and scale adaptative detection of small targets.Finally,since the residual regions in the salient map,a directional consistency check method based on the local isolation and the isotropy of the target is implemented to filter the candidate target regions,which can eliminate false targets and improve the robustness of the algorithm.(3)Experimental verification of algorithm performance.In order to verify the detection performance of the algorithm in this paper,a small target detection dataset of infrared sequence images under complex scenes is constructed for its comparison experiments with four baseline algorithms.The experimental results show that the algorithm in this paper has certain advantages over other baseline algorithm;it can achieve scale adaptation and precise localization of small targets,at the same time,its detection results are robust and reliable.It is proved that this method is a kind of infrared small target detection methods which can achieve a good compromise between improving detection performance and reducing false alarm rate.