TWC-AWT-Net: a transformer-based method for detecting ships in noisy SAR images

Synthetic Aperture Radar (SAR) plays an important role in marine ship management due to its all-weather, multi-band, and multi-polar characteristics, but the complex environment on the ground and the limitation of the imaging principle lead to the existence of large noise interference in SAR imaging, which poses a greater challenge to the existing ship detection algorithm. To address the problem of low detection rate and high false alarm rate of ship detection in noise-laden SAR images, we propose a detection algorithm based on convolutional neural network and transformer. First, we design the processing structure of multi-scale feature maps based on the advantages of transformer global processing capability, then we use weighted fusion of different scale feature maps after global processing to enhance the detection capability of the model for noise-containing targets, and finally, the model is evaluated using the high noise SAR datasets (HNSAR) we constructed. The experimental results show that our model has better detection of noise-containing SAR images compared with other deep learning models.