Transfer learning-based Traffic Identification for UAV-Assisted IoT

Abstract

Traffic datasets are costly and difficult to attain in the UAV-assisted IoT environment, and the time-sensitivity of traffic distribution is high, which makes it difficult for traditional machine learning traffic identification methods to be applied in practice. To address this challenge, we propose a transfer learning-based approach for UAV-assisted IoT traffic identification: TLB-CNN (Transfer Learning Based Convolutional Neural Network). Firstly, the initial model of the convolutional neural network is pretrained based on the source domain-complete IoT dataset, and then the retraining of the convolutional neural network is realized through the layer frozen based fine-tuning learning algorithm on the incomplete dataset in the target domain. The experimental results indicate that our method can effectively ensure the accuracy of traffic recognition under the conditions of limited traffic training samples. Compared with existing few-shot learning methods, the classification performance is significantly improved