Research on ResNet34 Improved Model

ResNet34 is one of the important models for image classification, and research has found that ResNet34 has problems such as information loss, low accuracy, and training difficulties. This paper proposes an algorithm based on improving ResNet34. Aiming at the problems of high depth and high complexity of ResNet34, an improvement scheme is proposed to lighten the model, speed up the training and convergence of the network, and make the algorithm more efficient. Aiming at the problem of ResNet34’s large computational volume and difficult training, the model is trained by using mixed precision and cosine annealing attenuation, which accelerates the convergence speed of the network and improves the classification accuracy of the model at the same time.