Cloud-Edge Collaboration Feature Extraction Framework in Satellite Multi-access Edge Computing

Mobile image recognition based on satellite multi-access edge computing can meet the needs of users in the coverage of nonterrestrial networks, and provide a great convenience for people's production and lives. Most existing methods extract features of the image data at the satellite servers, and only upload the extracted features to the cloud server for further processing, so as to reduce bandwidth pressure from satellite servers to cloud servers and response time. However, in most of these methods, the feature extraction on the satellite server is separated from the cloud server. This leads to the lack of effectiveness of the extracted features, which cannot meet the user's requirements for high accuracy of image recognition. To this end, we propose a cloud-edge collaboration feature extraction framework, in which the feature extraction on the low earth orbit satellite server is performed with the assistance of the cloud server. In the proposed framework, we first use the data set stored on the ground cloud server to generate a feature extractor E that learns how to extract effective features by exploring the distribution of the data set. Then, the cloud server sends extractor E to the satellite server to extract features of the data on the satellite server. In this way, extractor E can extract the effective features of the image data on the satellite server. Experimental results show that our method can achieve similar or even better recognition accuracy with uploading raw image data when only uploading a small amount of feature data. In addition, compared with existing methods, our method reduces the network traffic by about 98% in the case of similar recognition accuracy.