Wavelet Neural Networks Used In Oil Spill Remote Sensing Image Denoising

Wavelet neural network is a new type of neural network which is based on wavelet analysis, it has fully inherited from the both worlds of wavelet analysis and neural network. On one hand, multi-scale wavelet transform analyzes the signal by means of scale transform and translation transform that can effectively extract the information of local signals; on the other hand, neural network has the advantage such as self-learning, adaptive, fault tolerance and etc, and is also a universal function approximator. It has been proved theoretically that wavelet neural network has the ability of both uniform approximation, and L2 approximation, and a simple structure, fast convergence. It can be a good solution to the oil spill remote sensing image with difficult to effectively extract data from massive, redundant data, and with low processing speed, accuracy and reliability problems caused by multi-sensor, multi-platform, multi-temporal, multi-spec-tral of the massive oil spill remote sensing image data. It could be seen that, use the wavelet neural network to denoising oil spill remote sensing image undoubtedly is an effective way.Wavelet neural network used in oil spill remote sensing image denoising, summed up in this paper accomplished:studied the theory of wavelet neural networks and wavelet neural network-related algorithms, accomplished a wavelet neural network algorithm of wavelet-based framework; used the conjugate gradient method to optimize wavelet neural network then in matlab programmed related network algorithm; according to the noise characteristics of oil spill remote sensing satellite image, used the optimized wavelet neural networks for oil spill remote sensing image denoising, verified and analyzed the effectiveness of the algorithm.Experimental oil spill image and calculation data show that the denoising algorithm has obvious advantages, not only a good addition to denoisation, but also a good addition to maximization the preservation of image details, and the denoised image closer to the ideal real images.