Abstract: Fire image recognition is an important part of the fire detection research. With a continuous deep development of the artificial intelligence application technology, genetic algorithms and neural networks have also been applied to fire image recognition. Aiming at the problem that network structure is not easy to determine in fire image recognition based on genetic and neural networks, a fire image recognition algorithm based on genetic algorithms by the niche technology and neural networks is proposed. According to the characteristics of fire image recognition, the algorithm firstly establishes the multilayer front neural network model, determines the number of input and output level nodes, and undecides the number of the concealed levels and the conceal level nodes in the model, and then codes the network structure and weight value, threshold value, and trains the network model using genetic algorithms by the niche technology and the classic genetic algorithms. The experimental results show that the algorithm can reduce the number of evaluation generations effectively and speed up the training process. And finally we perform fire image recognition by the trained neural network model, and obtain better results.

Key words: niche technology, genetic algorithm, neural network, fire image recognition