A Novel Active Semisupervised Convolutional Neural Network Algorithm for SAR Image Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Remote Sensing

Background:

• Convolutional Neural Networks (CNNs) show promise in Synthetic Aperture Radar (SAR) object recognition.
• Limited labeled data significantly degrades CNN performance in SAR object recognition tasks.

Purpose of the Study:

• To develop a novel active semisupervised CNN algorithm for improved SAR object recognition.
• To address the challenge of insufficient labeled samples in CNN training for SAR imagery.

Main Methods:

• Active learning is employed to select the most informative samples from unlabeled data to augment the training set.
• A semisupervised approach is implemented by incorporating a new regularization term into the CNN loss function.
• This method maximizes the utilization of class probability information present in unlabeled samples.

Main Results:

• The proposed algorithm demonstrates effectiveness in SAR object recognition even with a scarcity of initial labeled samples.
• Experimental results on the MSTAR database validate the algorithm's performance.

Conclusions:

• The active semisupervised CNN approach effectively mitigates the performance degradation caused by limited labeled data in SAR object recognition.
• The integration of active learning and semisupervised techniques offers a robust solution for data-scarce scenarios in SAR image analysis.