ResNet10: A lightweight residual network for remote sensing image classification
Gong Jiaming, Wei Liu, Mengjie Pei, Chengchao Wu, Liufei Guo
Abstract
Remote sensing image classification is challenging due to its complex background and various types. Currently, the commonly used lightweight neural network still has the problem of excessive calculation and parameter amount, which makes cheap mobile devices with low computing power unable to quickly complete remote sensing image classification tasks. In response to this problem, we propose a lightweight residual neural network-ResNet10 that is more suitable for use on cheap mobile devices with low computing power. Experiments based on the UCM Land-Use public data set show that the classification accuracy of ResNet10 can reach 96.2%, the size of the model is one-tenth that of ResNetl8, and it can classify 105 remote sensing images per second. ResNet10 makes it possible for remote sensing image classification to be used on mobile devices.