Multi-class scene recognition based on codal module and neural network
Yin Wu, Wei Pan · 2011
In this paper, a new scene recognition system IMSI is proposed, which based on partially connected neural evolutionary model and codal module. This module innovatively applies a new unit called codal module to connect several partially connected neural networks together. IMSI is effective for the multi-class scene recognition without feature extraction and solves the neural network oversizing problem when the scene picture becomes larger. After about 1000 generations of evolution, the average correct rate of IMSI in multiclass scene recognition achieved 77.5%. Experimental results show that our system is better than ordinary scene recognition method, single category, and more convenient to use.