Research On The Generation Technology Of Low-light-level Scene Matching Base Map

The rapid development of cruise missiles makes them play an important role in futurehigh-tech wars. The combination of aerial imaging and scene matching brings cruisemissiles bright eyes. Since the scene matching base map and the real-time map arecaptured at different time, in different imaging conditions, using different detectors, fromdifferent visual angles, different maps appear entirely different in optical characteristics(gray, color, etc.), geometric characteristics (shape, size, etc.) and spatial characteristics(location, direction, etc.), even if the same object. Furthermore, noise and interferencebring differences. When cruise missiles fly at night, the real-time map is night visionimages, while the base map is composed of satellite images or aerial photographs acquiredduring daytime in better weather conditions. The difference between daytime images andnight vision images greatly impact on the accuracy of scene matching and guidance.Therefore, the low-light-level imaging simulation is of necessariness. By reconstructing theLLL imaging system, daytime images (base map) are accurately converted to night visionimages. LLL Scene Matching base map simulation technology is one of the effective waysto improve the scene matching guidance accuracy of cruise missiles at night.The most important characteristic of LLL images is poor signal-to-noise ratiocompared to daytime images. Characteristics of LLL images are analyzed at the beginningof this paper, such as histograms, time-and-space-domain noise, image correlation, richnessof gray level and image information. Based on the analysis of two-dimensional histograms,information capacity is proposed as an indicator parameter to describe the richness andlocal correlation of night Vision image information. These characteristics are of vitaldirective significance to the evaluation of LLL image simulation.Atmospheric propagation is necessary in optoelectronic imaging process. Buildingimage weather model is always difficult, since the atmosphere is a complex systemchangeing randomly and effected by natural factors. In this paper, the influence ofatmospheric propagation on image quality in optoelectronic imaging process is analyzed.Radiative transfer theory is used to describe the atmospheric absorption and scatteringproperties and the atmospheric imaging system transfer function model is advanced. Basedon the mathematical model, the corresponding computer simulation software is developed.And LOWTRAN7 is embeded as a separate module to simulate the energy attenuation ofimages and energy distribution change due to atmospheric propagation, which laid the foundation for LLL imaging simulation.Spectral reflectance properties are the fundamental reasons for distinguishing onescene from another, as well as the contrast changes between daytime images and nightvision images. Based on the ground spectral reflectance properties and operationgprinciples of ICCD, spectral distributions of daylight and night sky light are analyzed. Theinfluence of ground spectral reflectance properties on imaging process is deeply discussed.The contrast difference between daytime images and LLL images is studied. Spetralradiance response and spatial response of ICCD are analyzed. A series of unifying modelsof LLL scene simulation is builded under different night illumination conditions, includingthe integral imaging model, MTF chain model and noise model of ICCD.,In order to meet the practical requirement of LLL imaging simulation, a novel imagesegmentation method based on low-light-level image characteristics is presented in thispaper. The contribution of K-means clustering in image segmentation is discussed. Thewhole process of Canny algorithm is analyzed in details. The maximum entropy based ontwo-dimensional histogram is proposed to determine the high threshold of Cannyalgorithm. This adaptive approach improves the effect of Canny algorithm and alsoenhances the applicability and practicability. Mathematical morphology is used in the edgeimage processingto get better image segmentation results.Finally, based on the models analyzed and builded in the paper, a real-time LLL scenesynthesis system is developed, including computing kernel of LOWTRAN7, imagesegmentation and category, ICCD simulation. And the experimental results are given atlast.