Comparison of AlexNet and ResNet Models for Remote Sensing Image Recognition

Abstract

Remote sensing image recognition is an important direction in remote sensing data processing. Among them, deep neural networks have achieved results far beyond traditional methods on many challenging image datasets in remote sensing images. There are various network models for convolutional neural networks, for better remote sensing image recognition, this paper has selected the AlexNet model and ResNet model for comparison to select the network model with more accurate results. After running, it found that the accuracy is 0.6479 on the ResNet model and 0.6023 on the AlexNet model, the dataset performs better on the ResNet18 model, to further elucidate this result, it analysed the difference in accuracy between the two models. It was found that the use of techniques such as learning residuals, jump connections, and batch variance in the ResNet18 model reduced the problem of gradient vanishing during training, while the deeper layers and higher number of parameters also led to higher accuracy of the ResNet18 model.