多Agent系统在遥感图像目标识别中的一种应用

Application of MAS based Remote Sensing Image Object Recognition

遥感图像处理技术的发展无论对民用还是军用都有重要的意义。多Agent系统因其智能性和并行计算能力在各种领域都有了广泛的应用。针对遥感图像处理中对于自适应、快速识别目标的需要,提出将多Agent系统应用于遥感图像目标识别,设计并开发了一个分布式目标识别系统。它利用多个边缘检测算子Agent检测边缘,将检测结果序列送入神经网络Agent进行分类,并与纹理统计Agent的结果相融合,判决目标类别。实验结果表明,该系统对不同种类、不同分辨率的图像有一定的普适性,有较高的实用价值。对今后多Agent系统在图像处理中的研究和应用有很大的意义。

The technology of remote sensing image processing is very important to civil and military application. Multi - Agent systems have broad applications in many fields for its intelligence and parallel computing. To meet the requirement of self - adaptive and quick processing in remote sensing image object recognition, a distributed architecture based on multi - agent system running at the same time on different computers with remote sensing images is developed. All those agents which detect object＇s edges send information to a neural network system whose job is to identify the shapes detected by the agents. And a texture analysis agent is also designed which is used as NN server agent＇s assistant to make a decision. Experiment results show that the method is fit for practical applications. The design of the architecture plays an important role in the research and applications in the future.