基于自适应仿生网络的非特定物体在线识别

Adaptive Bio-network Based Online Non-specific Object Recognition

目前的物体识别算法框架大多是训练与识别过程相对独立,这就造成系统只对训练过的类别有识别能力,如果物体或者环境发生变化,识别将失败,这大大降低了系统的适应性、可扩展能力和实时性。针对这个问题,提出一种基于自适应仿生网络的识别算法,算法通过多层神经网络模拟人类记忆结构,将学习到的物体知识存储之网络节点,且网络是可在线拓展的,使得系统可在线学习新的陌生物体,具有自适应性。实验表明,提出的算法能有效地改变传统识别框架带来的局限,能在线识别非特定的多种物体。

Most of the current object recognition algorithm framework is a relatively independent training and recognition process which result in the system can only recognize the learned objects .If the object or environment changes , recognition will fail , it greatly reduces the system＇s adaptability , extensible ability and real-time performance .Aiming at this problem , a kind of recognition algorithm based on adaptive bio-network is proposed .This algorithm uses the multi-layer neural network to simulation human memory structure , which will store the learned objects to network nodes as knowledge storage , and the network can be expand online , which makes the system can be online learning new objects .Experimental results show that the proposed algorithm can effectively change the limitation of traditional recognition framework , and is capable of on-line recognition of non-specific various objects .