【作者】 陈哲；

【导师】 张佳岩；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2025， 硕士

【摘要】 星载遥感红外图像检测系统具有全天候、抗干扰等优点,在小目标检测领域具有独特的优势,被广泛应用。由于星载遥感红外图像进行监测的目标大都是运动速度较快的船舰、飞机等物体。如果目标检测速度较慢,就达不到对目标实时位置监测的目的。因此,星载遥感红外图像小目标高速检测技术研究就成了一个很重要的研究方向。在星载遥感红外图像小目标高速检测技术的研究中,除了研究检测方法本身的检测性能,还应该考虑到检测方法部署的硬件平台。本文以卫星拍摄星载遥感红外图像为处理对象,针对星载遥感红外图像小目标检测时对实时性的需求,同时考虑了卫星常搭载的FPGA硬件平台的情况,提出改进的星载遥感红外图像小目标高速检测方法。旨在保证星载遥感红外图像小目标高速检测方法具有良好的检测性能,同时具有在FPGA平台高速检测的潜力。本文首先分析现有的星载遥感红外图像小目标检测技术,分析发现传统的利用目标强度特性、梯度特性一类的方法具有易在硬件实现、能高速检测的优势,同时适合于云层、海面背景下的红外图像小目标检测,然后从梯度特性和强度特性融合检测出发,设计了基于局部梯度和阶梯对比度融合的检测方法。然后根据分析阶梯对比度(HTCM)方法的检测模型,研究改进针对梯度特性的检测方法,设计阶梯梯度检测模板,得到阶梯梯度检测方法。该方法具有良好的检测效果。接着将阶梯梯度检测方法与阶梯对比度(HTCM)方法进行融合。此外,本部分工作还研究了针对不同目标大小的检测方法改进,引入了多尺度检测,提高了方法的检测性能。最后,以目前卫星多搭载FPGA硬件平台的情况。首先对时空域滤波方法进行硬件设计,分析方法在硬件实现存在的问题及检测方法可改进的方向。提出将基于阶梯对比度和阶梯梯度融合的多尺度检测方法从逐像素检测改进为逐图像块检测。改进后发现检测方法存在检测小目标形貌不准确、模糊的问题。本文参考硬件FPGA的并行处理特点,提出错位检测方法。引入错位检测方法后,得到基于图像块的错位阶梯对比度和错位阶梯梯度融合的高速检测方法,同时提出使用改进的局部自适应阈值计算进行图像分割。最终基于图像块的错位阶梯对比度和错位阶梯梯度融合的高速检测方法具有良好的检测性能,同时大幅度提升了在FPGA硬件上高速检测的潜力。更多还原

【Abstract】 Space-borne remote sensing infrared image detection system has the advantages of all-weather and anti-interference,and is widely used in the field of small target detection.Most of the objects monitored by satellite remote sensing infrared images are ships,aircraft and other objects with fast moving speed.If the detection speed of the target is slow,the purpose of real-time location monitoring of the target cannot be achieved.Therefore,the research on high-speed detection technology of small targets in space-borne remote sensing infrared images has become a very important research direction.In the study of high-speed detection of small targets in space-borne remote sensing infrared images,the detection performance of the detection method itself should be considered,as well as the hardware platform deployed by the detection method.This paper takes satellite-borne remote sensing infrared image shooting as the processing object,aiming at the real-time requirement of small target detection of satellite-borne remote sensing infrared image,and considering the situation of FPGA hardware platform often carried by satellite,proposes an improved high-speed detection method of small target of satellite-borne remote sensing infrared image.The purpose is to ensure that the high-speed detection method of small targets in satellite-borne remote sensing infrared images has good detection performance,and has the potential of high-speed detection on FPGA platform.Firstly,this paper analyzes the existing small target detection technologies in space-borne remote sensing infrared images,and finds that the traditional methods using target intensity characteristics and gradient characteristics have the advantages of easy implementation in hardware and high-speed detection,and are suitable for small target detection in infrared images under the background of clouds and sea surface.Then,the fusion detection of gradient characteristics and intensity characteristics is started.A detection method based on local gradient and step contrast fusion is designed.Then,according to the detection model of HTCM method,the detection method for gradient characteristics is studied and improved,and the step gradient detection template is designed to obtain the step gradient detection method.This method has a good detection effect.Then the step gradient detection method and the step contrast(HTCM)method are fused.In addition,this part also studies the improvement of detection methods for different target sizes,and introduces multi-scale detection to improve the detection performance of the method.Finally,the current satellite is equipped with FPGA hardware platform.Firstly,the hardware design of the time-domain filtering method is carried out,and the problems existing in the hardware implementation of the method and the direction that the detection method can be improved are analyzed.The multi-scale detection method based on step contrast and step gradient fusion is improved from per-pixel detection to per-block detection.After the improvement,it is found that the detection method has some problems such as inaccurate and fuzzy morphology of small targets.Based on the parallel processing characteristics of hardware FPGA,this paper presents a dislocation detection method.After the introduction of the dislocation detection method,a high speed detection method based on the dislocation step contrast and dislocation step gradient fusion of image blocks is obtained,and an improved local adaptive threshold calculation is proposed for image segmentation.Finally,the high speed detection method based on image block mismatch step contrast and mismatch step gradient fusion has good detection performance,and greatly improves the potential of high speed detection on FPGA hardware.更多还原