A SAR Image Target Recognition Approach via Novel SSF-Net Models.

Area of Science:

• Radar systems engineering
• Artificial intelligence in remote sensing
• Signal processing for target recognition

Background:

• High-resolution radar systems necessitate advanced techniques for accurate target identification.
• Radar Automatic Target Recognition (RATR) is crucial for distinguishing targets in complex environments.
• Synthetic Aperture Radar (SAR) image analysis is a key area of research for improved target recognition.

Purpose of the Study:

• To develop an efficient and accurate method for Synthetic Aperture Radar (SAR) image recognition.
• To design a novel convolutional neural network (CNN) for Radar Automatic Target Recognition (RATR).
• To enhance the processing efficiency and real-time performance of SAR target classification.

Main Methods:

• A Sparse Data Feature Extraction (SDFE) module was designed to leverage SAR image characteristics.
• A new CNN, termed SSF-Net, was proposed, incorporating the SDFE module.
• Three classification strategies were implemented within SSF-Net: three Fully Connected (FC) layers, one FC layer, and Global Average Pooling (GAP), with the latter two offering improved efficiency.

Main Results:

• SSF-Net demonstrated robust performance on public SAR datasets (SAR-SOC and SAR-EOC-1).
• The highest recognition accuracies achieved were 99.55% on SAR-SOC and 99.50% on SAR-EOC-1.
• The proposed method showed a 1% improvement in accuracy over comparison methods on the SAR-EOC-1 dataset.

Conclusions:

• The SSF-Net, utilizing the SDFE module, offers a superior approach for SAR image recognition.
• The network's efficient classification methods (one FC layer and GAP) provide better real-time performance.
• SSF-Net achieves state-of-the-art accuracy and robustness in Radar Automatic Target Recognition (RATR).