Inversion of plasma sheath parameters for hypersonic targets based on optimization of attention mechanism

Aiming at the problems of small sample size and complex data features in the parameter inversion task of plasma sheath sheath for hypersonic targets, it is proposed to design and optimise a lighter and high-performance network structure by integrating the SE attention module into a one-dimensional convolutional neural network (1D-CNN) combined with migration learning. A parameter inversion model that balances accuracy and efficiency is constructed by introducing deeply separable convolution to optimise computational efficiency. The network model is trained on HTV-2 and RAMC datasets to compare and analyse the performance of the network model after the addition of the SE attention mechanism module, and the average relative error value and maximum average error of the optimised inversion are reduced on the test set. The experimental results verify the effectiveness of the 1D-CNN model optimised based on the attention mechanism in the parameter inversion of the plasma sheath jacket, which provides a theoretical basis and technical support for the study of parameter inversion of hypersonic plasma targets.