Research On The Technology Of Infrared Small Target Detection Based On Wavelet Anisotropic Diffusion Filtering

Infrared small target detection technology is always important research topic in infrared image processing field. It is not only used widely in military field,but also strongly demanded and with application prospects in the civilian fields, and applied to forest fire monitoring, which is regarded as a kind of new thinking. Based on infrared target detection in military field, how was forest fire point regarded as infrared target is researched. Therefore, the main content of this dissertation is as following:Firstly, infrared radiation characteristic of infrared target(fire point) and background(forest background) is researched. Fire point of high temperature and non fire point of high temperature were emphatically studied in the monitoring of forest fire disaster. An important characteristic of forest vegetation is that its temperature varies in little scope, so it was made use fully in forest fire point detection, experimental result shows that detection rate is improved.Secondly, anisodiffusion and the undecimated wavelet transform(UDWT) were researched importantly. Due to the loss of thetranslation-invariance property in the decimated wavelet transform(DWT) , it leads to artifacts( pseudo-Gibbs phenomena) when an image is reconstructed. However, the research of image De-Noise shows that UDWT can inhibit effectively this phenomenon. The classical models of anisodiffusion were also researched. With anisodiffusion in wavelet domain,the target information becomes clearer in each frequency branch and three various directions of each frequency branch,and it was new detection method that is presented- detection method based on wavelets-anisodiffusion filter.Then, outdoor experiments were carried out by detectiong buried exception objects under ground with infrared technology(small fire or high temperature objects),and fire of ground surface and buried object were detected. In addition, visible and thermal imagery were obtained so as to analyze difference of them.Finally, forest fire point was detected by regarding it as infrared small target and on the foundation of the above research, wavelets-anisodiffusion filter forest fire point detection was presented. The classical forest fire model was studied to compare with former, which was designed for MODIS. However, data of Huanjing-Jianzai was used in this dissertation, which was launched by China in 2008, and the space resolution of HJ is 150m. Because of difference imaging waveband of them, fire threshold of classical model was modified and then it was applied to HJ.