Research On Key Technologies Of High Frame Rate Infrared Coded Imaging Based On Unmanned Platforms

With its low cost and continuous action,the unmanned platform can be used for tasks such as decision-making,search and rescue,strike,and surveillance.It is widely used in many fields such as sea,land,and air.Signal acquisition is the precondition and foundation for unmanned platforms to perform certain tasks.Therefore,all kinds of optoelectronic system have become an integral part of unmanned motion platforms.Because of the advantages of full-time work,strong environmental adaptability,and long-distance range,Infrared imaging is applied to unmanned motion platforms.With the further expansion of the scope of application,the infrared imaging are increasingly demanding imaging frame rates.The existing imaging systems have been difficult to monitor the needs of high-speed moving targets.For example,when detecting complex background drones,it is difficult to acquire minute detail transformations between infrared radiation of adjacent target scenes.But due to low imaging frame rate,resulting in low detection efficiency and limiting the Infrared imaging system application scenario.Therefore,this paper focuses on the key technologies of infrared high frame rate encoding imaging for unmanned platforms,focusing on infrared high frame rate imaging and infrared image quality improvement.The main research contents and achievements are as follows.(1)Considering the problem of low frame rate of traditional infrared camera imaging,this paper is studied an infrared high frame rate imaging technology based on encoding measurement.Firstly,the infrared radiation of the target scene is focused on the target surface of the light splitting module by using a front infrared lens,and the optical path is divided into two paths A and B.Secondly,a reflective spatial light modulator designing a code matrix is used for the light path A code modulation,through the infrared array detector 1 delay exposure,obtained after the modulation superposition of the encoding information,while,using infrared array detector 2 on the optical path B delay exposure to obtain the superimposed side information.Finally,using two-step iterative shrinkage threshold method reconstructs two sets of measured values to obtain multi-frame infrared images.Experimental results shown that the technology can effectively improve the infrared imaging frame rate and achieve good imaging results.(2)Considering the problem of generation of "ghosting" and low quality after correction for the traditional infrared image non-uniformity correction methods,this paper proposed a high fidelity non-uniformity method based on kernel regression.Firstly,the traditional neural network correction method was fully studied,and the expected image was reconstructed instead of the mean filter to obtain the predicted image.Secondly,the steepest gradient descent method was used to select the appropriate iteration step and update the gain and offset.Finally,the update the parameters was used to obtaining the corrected image.Experimental results shown that the proposed method can effectively suppress "ghosting" and improve image quality.(3)Considering the infrared image have low resolution and unclear details with high background and low contrast,this paper achieved an infrared image super-resolution based on the improved convolutional neural network by changing the activation function.Firstly,analysis of convolutional neural networks deeply,constructing a three-layer network model based on convolutional feature extraction layer,nonlinear mapping layer and image reconstruction layer.Secondly,using high-low resolution image pair training network to obtain network parameters.Finally,Input the low-resolution infrared image can output the final high-resolution image through the pre-trained network parameters,it can be realize the end-to-end super-resolution reconstruction.Experimental results shown that the proposed method can effectively improve image resolution and enhance texture detail information.