Research And Implementation Of Ear Recognition Based On Few-shot Learning

Biometrics refers to the technique of distinguishing identities by using computer methods combined with optical,acoustic,biosensor and biostatistics.There are many sub-fields,and human ear recognition is one of the important branches.Due to its unique structural characteristics and position,its theoretical and applied research has attracted more and more attention from researchers at home and abroad in recent years.Deep learning is a technology that has been applied in the field of image recognition in recent years.Convolutional neural network(CNN)is one of the most widely used types of networks,and has achieved excellent results in image recognition.However,deep learning requires a large amount of data to train neural networks in order to obtain high recognition accuracy,while the human ear field currently lacks large-scale public datasets.Based on the cutting-edge few-shot learning technology,this paper uses TensorFlow,a deep learning framework,to design a convolutional neural network for human ear recognition research,which improves the accuracy and robustness of human ear recognition.The main research contents are as follows:(1)A human ear recognition network was designed.Firstly,the structural characteristics of the convolutional neural network are introduced and analyzed.Then,the layer-by-layer design of each layer is carried out based on the key structure of the convolutional neural network.Then the activation function and dropout regularization are optimized.Finally,the whole network detailed design is given.(2)Aiming at the lack of data in the human ear dataset,a human ear recognition method based on meta-learning algorithms for few-shot learning is proposed.The TensorFlow framework is used to determine the optimal interval of learning rate and the optimal value of dropout probability.Finally,the experimental results show that the proposed method has better accuracy and robustness than the traditional transfer learning based fine-tuned network.(3)Based on the proposed network model and method,a few-shot learning based human ear recognition application is designed and implemented.The overall structure is firstly designed,and the identity information collection module,human ear image preprocessing module,the image recognition module and the human ear image recognition result processing module were designed in detail,and implemented and tested at the end.