Study Of Multisensor Data Fusion Algorithms

Study and application of multisensor data fusion theory have aimed to combine multiple source information from various sensors intelligently and obtain more detailed, complete, dependable description and decision than a single sensor. Application of multisensor data fusion span a broad range that includes military and civilian. Multisensor data fusion is one popular research field in domestic and foreign scientific and technological circles. This dissertation deeply deals with multisensor image fusion, analysis of colored texture image, radar network detection system around the three levels of data fusion: data, feature, decision. A lot of new ideas and approaches are proposed and better results are achieved. The main contributions in this dissertation can be summarized as follows: * A new method is developed to merge a high-resolution panchromatic image and a low-resolution, multi-spectral image based on à trous algorithm. Firstly this method presents a sort of fusion idea of region division based on à trous multi-resolution wavelet decomposition. Then, the evolutionary strategy is used to solve successfully the problem how to select threshold in dividing region. Finally, an influence fusion factor is proposed and verified. The fusion results show that this method can perform better than the IHS and à trous wavelet merger methods.* A new technique is presented based on Mallat algorithm for the fusion of a high-resolution panchromatic image and a low-resolution, multi-spectral image. This method can effectively merge the high-resolution panchromatic image approximation into the multi-spectral image approximation using fusion factor by means of multi-resolution wavelet decomposition, which can make an improvement on Mallat wavelet merger method.* Image fusion methods base on Laplacian pyramid decomposition and Mallat algorithm are discussed in detail, and their shortcomings are pointed out. A new method is developed to merge two spatially registered images with differing focus points based on multi-resolution wavelet decomposition and evolutionary strategy. This method has the advantage that redundant information at multi-scale and in multi-direction can be fully used by means of shift-invariant wavelet decomposition and ringing from the final merged image reconstructed by Mallat algorithm can be overcome. Color space transforms are studied based on gray texture analysis. A new representation for color texture is proposed by effectively merging both the texture and color information in feature-level based on incomplete tree-structured wavelet decomposition(ICTSWD), which can* describe color-texture feature with more exact and complete. Feature-level fusion and classification can be performed on the basis of the pyramid wavelet decomposition(PWD), ICTSWD and wavelet packet decomposition (WPD). The results demonstrate that colored texture feature based on ICTSWD has better classification performance and anti-noise ability than other features based on PWD and WPD. * In radar network system, wavelet filter based on soft-threshold is combined with the parallel distributed detection fusion system with multiple sensors ideally. A theoretical analysis is presented in the sense of the Neyman-Pearson (N-P) test about the relationship between fusion rule and local decision rules in the parallel distributed detection fusion. The simulation results show that synthesize algorithm can enhance the radar detection performance remarkably.