A software engineering approach to develop adaptive RBF neural networks

Component-based software engineering is away of raising the level of abstraction of software development so that software can be built out of existing context -independent components that can be widely reused. Research has shown that component-based software engineering leads to software that is of higher quality, is developed in a shorter time and therefore results in lower cost. However, the lack of a framework for expressing component collaboration makes component-oriented programs more difficult to maintain, expand and widely reuse. This paper demonstrates the use of an adaptive component-based meta model driven framework that eases the integration of heterogeneous components into an application at runtime. By using the proposed framework, this paper introduces a new software engineering approach to the implementation of an adaptive RBF network. The dynamic RBF network is applied to an image restoration task in order to realize the functional mapping from a degraded image space to the original image space, where no prior knowledge and assumptions about the blurring process and the additive noise are required. The proposed RBF network can run in either sequential or parallel modes.