研究目的
Investigating the feasibility and efficiency of a deep learning-based method for real-time atmospheric refractivity inversion to support ship communication, navigation, and radar detection.
研究成果
The MLP-based atmospheric refractivity inversion method demonstrates feasibility and efficiency, providing accurate inversion results in less computation time compared to the PSO algorithm. This method offers a strong support for the performance prediction and evaluation of radar and communication systems by obtaining the distribution of atmospheric ducts over the sea in real time.
研究不足
The study is based on simulated data sets, and the validations of the deep learning-based refractivity inversion method in practical applications will need to be tested by real data experiments in future investigations.
1：Experimental Design and Method Selection:
A multilayer perceptron (MLP) with five hidden layers and rectified linear unit (ReLU) activation function is chosen for the inversion method. The mean-squared error is used as the loss function, and the adaptive moment estimation (Adam) algorithm is selected for training.
2：Sample Selection and Data Sources:
A pregenerated database of 100,000 environments with duct parameters values generated using three-dimensional Latin hypercube sampling. Propagation losses are computed using the split-step Fourier parabolic equation method.
3：List of Experimental Equipment and Materials:
MATLAB 2016a on a computer with an Intel i5 3.2 GHz CPU.
4：2 GHz CPU.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: The MLP is trained with the database to invert the refractivity profile. The batch size is 100, and 400 epochs are used. The dataset is divided into 80% for training, 10% for validation, and 10% for testing.
5：Data Analysis Methods:
The inversion results are compared with those obtained from the particle swarm optimization (PSO) algorithm to validate the effectiveness and accuracy of the MLP-based method.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
