Research on Cat and Dog Image Recognition Based on Several Classic Neural Networks

This article uses TensorFlow and Keras API tools to build a neural network, and compares and analyzes the recognition rates of several classic neural networks (LeNet, AlexNet, VGGNet, GoogLeNet, ResNet) on cat and dog picture sets. The ResN et network uses transfer learning methods to accelerate The training speed of the neural network, the final results show that the training accuracy of the LeNet neural network is about 65%, and the verification accuracy is about 62%. AlexNet neural network training accuracy is approximately 92<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, and validation accuracy is approximately 83%. The VGGNet neural network training accuracy is approximately 94<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, and the verification accuracy is approximately 91 %. The GoogLeNet neural network training accuracy is approximately 99%, and the verification accuracy is approximately 97%. ResNet running results, we can know from Figure 1 that the training accuracy is approximately 97% and the verification accuracy is approximately 95%. Judging from the above results, GoogLeN et and ResN et neural networks have the same accuracy as humans in identifying cat and dog images, and have wide application value.