Research On Key Technologies Of Ship Target On-orbit Detection In Infrared Remote Sensing Images

Efficient on-orbit remote sensing image processing and detection techniques are of great significance for extending the scope and improving the efficiency of remote sensing applications.This work is based on the development of large-scale highresolution spatial infrared remote sensing and belongs to the frontier technology.Main works are summarized as follows.1)This paper comprehensively investigated and analyzed the ship’s imaging mechanism,imaging characteristics,sea surface temperature distribution,and changes.Besides,the requirements of the detector band,imaging mode,and sensitivity were analyzed.On this basis,the data acquisition subsystem was developed to solve the data source problem of algorithm verification.Aiming at the characteristics of data acquisition,the complete processing flow and hardware framework from image preprocessing to ship target detection were proposed.2)In terms of image preprocessing,this paper studied the nonuniformity correction and blind pixel processing techniques to improve image quality.In the nonuniformity part,a correction method based on curvature regularization was proposed.From the variation point of view,the introduction of improved curvature regularization reduced the amount of computation and preserved the gradient direction for the characteristics.The method effectively improved the stripe nonuniformity in the image,and each index is superior to the traditional correction method.In the blind element part,an improved blind pixel detection method for directional information measurement nonlinear diffusion is proposed.Based on the non-linear diffusion model of direction information measurement,the concept of confidence was used to achieve more accurate blind element detection.The proposed method not only effectively removes blind elements but also retains more details.3)In terms of ship target detection,this paper studied the traditional method and deep learning method separately.In the traditional method part,a method based on saliency in infrared remote sensing ship target detection was proposed.The saliency method and the stability factor was used to enhance the attention of the target area,which could effectively detect the ship target.In the deep learning part,a ship detection network with a lightweight Encoder-Decoder structure was proposed.The method combined the advantages of a binary neural network and semantic segmentation network to reduce the consumption of storage and computing resources,and achieve fast and efficient detection results.In the process,the influence of image quality and pretreatment effect on the detection efficiency was studied,which lays a foundation for the practical application of processing methods and algorithms.4)In terms of online implementation of software and hardware,for the future engineering application requirements,the implementation of algorithms based on lowpower hardware platforms is carried out.The hardware overall framework is designed,the resource requirements of the hardware of the detection method are analyzed,and the embedded software development based on the lightweight coding-decoding network is completed.The function of target detection is realized,and the effectiveness of the detection method and online implementation are verified and feasibility.This article is a preliminary technical exploration of the overall work.The key technologies of the real-time on-orbit ship detection process algorithms and hardware implementation of space infrared remote sensing were studied.Its feasibility has been initially verified.The research content of this paper has important academic and application value,which lays a foundation for further engineering application.