Scholarlink

<p>Although traditional GNSS positioning methods such as Kalman filtering and least squares have been proven to be effective in improving GNSS positioning accuracy, their actual positioning results are highly dependent on assumptions regarding the GNSS observation noise and motion models, thus, these methods are difficult to apply in different environments. Therefore, a temporal convolutional network model is proposed based on a squeeze-and-excitation module and multi-scale feature fusion for GNSS positioning error correction. The squeeze-and-excitation module is an attention mechanism used to improve the performance of convolutional neural networks, which can better capture key features in the data. Multi-scale feature fusion helps the model better understand and utilize features of different scales, thereby improving the performance of the model. The proposed scheme first preprocesses the input data. Then the approximate position of the receiver is estimated using the Kalman filter method and fed into the proposed model as prior information along with features extracted from GNSS observation data. Subsequently, the presented model is adopted to perform error correction on the receiver position. Finally, the model outputs the corrected longitude and latitude of the receiver. The dataset used in the experiment consists of GNSS data such as satellite positions, raw pseudoranges, and carrier-to-noise ratios collected by multiple low-cost receivers in the San Francisco area of the United States. The environments where GNSS data was collected include highways, tree-lined streets, and urban canyons. The data in the training set and the test set were collected separately, and the receiver was moving during the data collection process. Experimental results show that the proposed method not only outperforms the least squares, Kalman filter and long short-term memory network in positioning accuracy in three different environments, but also improves the overall positioning accuracy by 28.8% and 4.7%, respectively, compared to the current state-of-the-art graph convolutional neural network methods and reinforcement learning methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Temporal convolutional network based on squeeze-and-excitation module and multi-scale feature fusion for GNSS positioning error correction

  • 期刊论文

  • 发表日期：

    2025年6月14日

查看全文
  • Xiuxun Liu,
  • Zuping Tang,
  • Jiaolong Wei

摘要

Although traditional GNSS positioning methods such as Kalman filtering and least squares have been proven to be effective in improving GNSS positioning accuracy, their actual positioning results are highly dependent on assumptions regarding the GNSS observation noise and motion models, thus, these methods are difficult to apply in different environments. Therefore, a temporal convolutional network model is proposed based on a squeeze-and-excitation module and multi-scale feature fusion for GNSS positioning error correction. The squeeze-and-excitation module is an attention mechanism used to improve the performance of convolutional neural networks, which can better capture key features in the data. Multi-scale feature fusion helps the model better understand and utilize features of different scales, thereby improving the performance of the model. The proposed scheme first preprocesses the input data. Then the approximate position of the receiver is estimated using the Kalman filter method and fed into the proposed model as prior information along with features extracted from GNSS observation data. Subsequently, the presented model is adopted to perform error correction on the receiver position. Finally, the model outputs the corrected longitude and latitude of the receiver. The dataset used in the experiment consists of GNSS data such as satellite positions, raw pseudoranges, and carrier-to-noise ratios collected by multiple low-cost receivers in the San Francisco area of the United States. The environments where GNSS data was collected include highways, tree-lined streets, and urban canyons. The data in the training set and the test set were collected separately, and the receiver was moving during the data collection process. Experimental results show that the proposed method not only outperforms the least squares, Kalman filter and long short-term memory network in positioning accuracy in three different environments, but also improves the overall positioning accuracy by 28.8% and 4.7%, respectively, compared to the current state-of-the-art graph convolutional neural network methods and reinforcement learning methods.