Study On Multispectral Image Registration Methods In Forest Monitoring

Forests are a valuable natural resource, which covers approximately9.4%of the Earth’s surface area or,30%of the surface area of land, and is of great significance for global balance of ecology and carbon cycle. With the rapid economic development in the recent hundreds of years, there was an increasing demand for wood, resulting in large areas of’deforestation, and extinction or endangering of many species. The frequent occurrence of forest fires also caused significant damage to forests, making forest protection extremely urgent. Due to the large volume and wide coverage of forest, it is relatively difficult to realize comprehensive monitoring. But with the continuous improvement of monitoring and imaging equipment, satellite monitoring had become an important means of forest monitoring. Satellite multispectral images included comprehensive information of the forest, but a lot of such information cannot be observed with the naked eye without processing and calculation. This paper mainly focused on the studies of registration of satellite multispectral image.In this paper, the concept, principles and methods of image registration were outlined, and the different transformation of general pictures was achieved using MATLAB software. With the introduction of the application of feature-based method and gray-based method in image registration, a multispectral image registration method was derived from the former methods. By comparing the registration effect of SIFT and mutual-information method, the experiment showed that the former method was better for dealing with ordinary images, but not appropriate for multispectral remote sensing images. By using mutual information method for multispectral image registration, the experimental results showed that the traditional mutual-information method had a good effect on entire pixel shifting, but still needed improving in sub-pixel shifting. Firstly, integer linear interpolation and down sampling image Fourier transformation theory were employed to testify the accuracy of the method; then the sub-pixel registration process and theoretical analysis was shown; Finally, through experimental verification of different bands of sub-pixel images’movement, the results showed that the highest point of mutual information was the theoretical matching point, and also the best matching point.Sub-pixel accuracy registration required Powell and other algorithms, but the sub-pixel accuracy of the registration was not high, with fluctuating performance surfaces. It was difficult to search the maximum point in an optimized way, with the traversal scan process being time-consuming and tedious. Due to the approximation of the mutual information’s performance surface to Gaussian distribution, this paper proposed a new sub-pixel image registration method based on Gaussian model. With formula derivation and experimental demonstration, the results showed that the method was featuring convenient calculating, superior reliability and high accuracy in multispectral image registration, which reached the sub-pixel level.