一种新的模糊多层自组织神经网络图像目标检测方法

A Novel Target Detection Approach in Images by Fuzzy Multilayer Self-organizing Neural Netwok

神经网络方法和模糊集理论用于图像处理和目标检测时各有优势,文中提出了一种将神经网络和模糊集理论集成的混合方法,即模糊多层自组织神经网络法。该方法将模糊测度作为神经网络的目标函数,网络包括多层结构,任一层中的一个神经元对应图像中的一个像素,该神经元只与前一层的对应元素及其邻域元素连接。针对遥感图像的实验处理过程证明该方法能够有效地进行目标检测和提取,并且具有良好的噪声免疫力。

There are respective advantages for image processing and target detection by neural network methods of fuzzy set theory methods.The paper proposes a hybrid approach by combining the two type methods,namely,Fuzzy Multilayer SelfOrganizing Map(FMSOM),which utilizes fuzziness measure as objective function of neural network.The network comprises multilayer structure,every neuron in each layer corresponds to pixel of image,which just connects with the corresponding neurons in prior layer and its neighbors.Experiments using remote sensing images as input are executed and the results verify the approach is valid for target detection and extraction,simultaneously,possesses good noise immunity.