GNSS-Based Scene Recognition by Means of Machine Learning

Scene recognition has attracted considerable attention in the field of navigation in recent years since the signals can be used for navigation depending on the specific scene in which the receiver is located. Scene recognition can assist in developing a scene-adaptive navigation system that selects different positioning algorithms or sensors for navigation, enabling accurate positioning and navigation in different environments. In this study, we propose a supervised machine learning model. Based on the multi-satellite GNSS observation, 16 features are designed, and the most effective 8 features are selected from them as model inputs through the Chi-square test. Support Vector Machine and Random Forest are utilized as classification models with the strategy of result fusion to divide the outdoor navigation scenes into four types: Highway, Suburb, Urban Canyon, and Avenue. The test results show that our proposed scene recognition algorithm achieves an overall recognition accuracy of 94.37%, with each of the four types of scenes having a recognition accuracy exceeding 93%, which shows better performances than other existing methods.