원문정보
초록
영어
Aiming at the complexity and limitations of traditional character recognition design method, an algorithm combined with genetic algorithms and neural network is proposed. Using this method, the advantages genetic algorithm which global optimal solution or a very good performance suboptimal solutions can easily be obtained is fully utilized. The shortcomings of neural network model such as slow convergence speed, entrapment in local optimum, unstable network structure etc are solved. Combined neural network and genetic algorithm is to make full use of the advantages of both, so that the new algorithms both neural network learning capability and robustness, but also a strong genetic algorithms global random search capability, the neural network has self-evolutionary, adaptive capacity, so as to construct evolutionary neural network. The actual application in character recognition results show that, compared with the traditional method, this model has a strong feasibility and effectiveness.
목차
1. Introduction
2. Handwritten Digit Symbol Image Preprocessing
3. Character Feature Extraction
3.1. Cross-cut Features
3.2. Projection Characteristics
3.3. Structural o f Feature
3.4. Coarse Grid Characteristics
4. Character Feature Extraction
4.1. Determine the Network Layers
4.2. Input and Output Mode
4.3. BP Network Learning Rates and Learning Algorithms
4.4. Activation Function
4.5. Hidden Layer Nodes
5. Training and Recognition of the Network Model
6. Conclusion
Acknowledgements
References