Multi-scene Infrared Dim Small Target Low False Alarm Detection Algorithm For High Speed Mobile Platform

Infrared dim small target Low false alarm detection algorithm is the key technology of infrared long-range detection system for high-speed mobile platforms such as fighters,cruise missiles and ballistic missiles.This kind of platform will change the scene when detecting long-range targets.According to Neyman-Pearson criterion,it is difficult to have a set of algorithms that can take into account the detection rate and false alarm rate of all scenarios.The detection threshold of the algorithm and the noise suppression of the filter are often suitable for a certain scene or a certain clutter.To obtain higher detection rate,the detection threshold needs to be lowered,but once the background clutter and noise in complex scenes exceed the current threshold,False alarms will be generated.In view of the above analysis,this paper proposes a multi-algorithm parallel operation real-time infrared dim small target detection algorithm signal processing flow and its system framework.Firstly,this paper analyzes and studies the imaging mechanism and image characteristics of stationary ground-based platforms for infrared dim and small targets in the air and infrared dim and small targets facing mobile platforms,and simultaneously constructs the infrared multi-scene target and background datasets of high-speed mobile platforms required in this study.,which were divided into dim and small target detection datasets and multi-scene and multi-complexity background datasets through different annotation methods.In order to improve the signal-to-noise ratio of the target and enhance the background texture features,a multi-layer progressive image processing algorithm was designed,including:suppressed and filled the defect pixels in the infrared imaging system based on the statistical characteristics of space-time;Combined the detection system and information entropy to achieve real-time focusing of the target area,which improves the intensity of weak and small targets from the system perspective;and the dual-channel image enhancement algorithm was used to quantify the detail features of the map width to provide a foundation for subsequent scene classification.Then,the characteristics of imaging system noise and false alarms in different scenarios were introduced and analyzed.According to the complex and fast-changing scene presented by the infrared imaging background under the mobile platform and the hardware implementation requirements of the signal processing system,and drawing on the idea of Multi-Task Learning,a real-time infrared scene classification and segmentation multi-task learning network based on bilateral fusion was proposed.Model（BFM-SCSeNet）,this lightweight network model accurately and effectively outputs the full-image scene classification and region segmentation results of the current frame infrared image in parallel and real-time.In order to reduce the false alarm rate of infrared dim target detection algorithm,On the premise of obtaining the region segmentation results of the previous frame of image,combined with multi-aspect ratio saliency and guided filtering technology,an infrared dim and small target detection algorithm based on subregional adaptive constant false alarm was proposed,which was suitable for The complex cloud background and the simple empty background in the air have a high robust detection ability to the target shake.At the same time,the characteristics of false alarms appearing in the complex background of the underground view were fully analyzed,with the original intention of minimizing false alarms,a frame skipping mechanism based on the relative speed-to-high ratio of high-speed platforms was proposed,and a frame combining KLT and nuclear correlation filtering was formed.Inter-difference infrared dim and small target detection algorithm.At the end of the algorithm,the vector motion relationship between possible false alarms was used to finally eradicate the interference of complex ground false alarms.Finally,according to the algorithm idea and parallel signal processing flow of"according to the suit,the right medicine",using the VPX architecture,a set of hardware system framework for parallel execution of multi-algorithms for low false alarm detection of infrared dim and small targets in multiple scenes was proposed:The laminar water signal processing framework PMPSPF,the circuit design of its main hardware modules,and the engineering implementation effect of the framework were analyzed.The experimental results showed that the local average signal-to-noise ratio of the target is greater than 0.89d B,and the target scale is greater than 1 pixel.Infrared targets of multiple scales and scenes were detected by hardware system.The average detection rate of different scenes reached 99.3%,and the false alarm rate was controlled below 10^-5,which verifies the correctness of the algorithm proposed in this paper and the proposed hardware.The practicability of the processing architecture provided a set of practical solutions for low false alarm detection of multi-scene infrared dim and small targets for high-speed maneuvering platforms.