| With the development of pattern recognition,the amount of face recognition data increases sharply,which requires the model to have higher recognition accuracy and efficiency.In order to effectively improve the efficiency of the model under large samples and at the same time improve the problem of the recognition accuracy decline caused by misclassification and misclassification,this thesis introduces the attention mechanism with high information processing ability and the support vector machine with high robustness into the convolutional neural network to build a novel deep learning face recognition classification model,effectively extract the key information in the image and deal with the noise data.The proposed model can significantly improve the accuracy and efficiency of face recognition and provide a new idea for the research of face recognition and pattern recognition.The main work of this thesis is as follows:1.Selection of activation function.In this thesis,the advantages and disadvantages of each function are analyzed,so as to select the most suitable function for the model,and the convolutional neural network model is derived according to the logic structure and working principle of feedforward neural network.At the same time,the preprocessing simulation experiment of images with different interference factors has laid a foundation for the follow-up work.2.Introduction of attention mechanism and support vector machine.By introducing the attention mechanism to calculate the distribution of attention,the problem of gradient explosion and the decline of computing power caused by too much information are avoided,so the processing power of information is improved.By introducing support vector machine as classifier,convolutional neural network model output as input to find the maximum classification interval hyperplane,in order to achieve the optimal classification results,and improve the performance of the model.3.Compare the basic convolutional neural network model with the improved model and other optimized models on multiple face data sets.The results show that compared with the basic model,the recognition accuracy of the improved model on the data set is increased by2.5% and 3.3% respectively.In order to verify the universality of the new model,a comparative experiment was conducted on a self-built data set with a capacity of 10000.The results show that the recognition accuracy of the improved model on the self-built data set can reach 99.5%,which is 2.3% higher than that before the model improvement,and the new model has higher convergence efficiency on each data set,which confirms the feasibility of the proposed method. |
