Study On Target Detection Algorithms And Its Simulation Based On Synthetic Aperture Radar And Optical Images Fusion

Recently, image processing on Synthetic Aperture Radar(SAR) hasbecome one of the research focuses, due to its high efficiency in militaryfield and remote sensing.However, under the requirement of all-time-all-weather target detection and recognition, a certain type of sensor, SAR oroptical sensor, could not fulfill the practical need in some specificenvironments. Thus, further research on multi-sensor image fusion has beenwidely discussed.Based on practical project, this thesis focused on the study of imagedenoise, target detection and multi-sensor image fusion, and realizes thevisual simulating platform for target detection using SAR and optical multi-sensor image fusion. The main work in the thesis is as follows:Firstly, based on the characteristic of the speckle noise and the need ofboth enhancing the wavelet-denoise and overcoming the shift invariant ofdenoise in CT transform, a new denoise method is brought out, using NSCTtransform with self-adaptive shrinking coefficients. Then a comparebetween the results and those using traditional methods is provided.According to the experiments, it is proved that the NSCT denoise methodsperforms better for its higher maintaining of boundary information.Secondly, based on the characteristics of SAR and optical images, anew target detection solution using the information of boundaries wasbrought out. When detecting the boundaries, by searching linking points inrecursion, the broken part of a boundary could find its nearest boundary andmerge with each other. When determining whether a certain region belongsto target, the decisions are made by calculating its trust degree. Accordingto the experiments, it is proved that this method decreases the dependency dependency on the accuracy of thresholds, also bears a high efficiency andthe accuracy of image segmentation could be enhanced greatly.Moreover, to overcome the limitation of registering SAR and opticalimages, a new fusion strategy is put out to enable the multi-sensordetection based on image fusion under the situation of low accuracy ofimage registration. Nearby regions in two images are fused based on thefussy theory.According to the experiments, it is proved that this strategydecreases the requirement of accurate registration, gets a more accurateand completed information of target than single sensor does.At last, according to the project, a high performance integrated systemof multi-sensor target detection is provided based on MATLAB-GUI,which offers an experimental platform for the multi-sensor images denoise,enhancement, boundary detection, and fusion. Taking SAR and opticalimages for example, the structure of the system and all functions arerealized. It builds a solid foundation for further researches in this field.