SAR ship target detection method based on CNN structure with wavelet and attention mechanism.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Ship target detection in Synthetic Aperture Radar (SAR) images is crucial but challenging due to sea clutter and noise.
• Existing deep semantic segmentation networks often fail to fully utilize global image information for SAR ship detection.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) method, named WA-CNN, that integrates wavelet transform and attention mechanisms for improved SAR ship detection.
• To enhance the utilization of global image information and preserve target details in noisy SAR images.

Main Methods:

• The proposed WA-CNN utilizes a U-Net structure with an encoder-decoder architecture.
• Dual Tree Complex Wavelet Transform (DTCWT) is incorporated into the encoder to reduce speckle noise and preserve target contours.
• An attention mechanism is integrated into the decoder to capture global contextual information of ship targets.

Main Results:

• The WA-CNN method demonstrated effective performance in ship target detection using two public SAR image datasets.
• Experimental results indicate that the proposed method successfully mitigates noise and enhances the detection of ship targets in complex SAR imagery.
• The integration of DTCWT and attention mechanisms improved the network's ability to utilize global information and preserve target details.

Conclusions:

• The developed WA-CNN algorithm is an effective and feasible approach for ship target detection in SAR images.
• The method shows significant potential for applications requiring accurate and robust ship detection in challenging marine environments.
• The study highlights the benefits of combining wavelet transforms and attention mechanisms for processing noisy SAR data.