High-Resolution Radar Target Recognition via Inception-Based VGG (IVGG) Networks.

Area of Science:

• Radar systems engineering
• Artificial intelligence
• Machine learning

Background:

• Accurate radar target recognition is crucial for defense and surveillance.
• Existing convolutional neural networks (CNNs) face challenges in processing complex radar signals like HRRP and SAR.
• The Visual Geometry Group (VGG) network offers a robust architecture but can be computationally intensive.

Purpose of the Study:

• To develop novel CNNs, termed Inception-based VGG (IVGG) networks, for enhanced radar target recognition.
• To improve the efficiency and accuracy of target classification using HRRP and SAR data.
• To adapt and optimize deep learning architectures for the specific characteristics of radar signals.

Main Methods:

• Introduction of the Inception module into the VGG network architecture.
• Modification of the full connection layer's connection mode.
• Addition of a pointwise convolutional layer to enhance network nonlinearity.
• Comparative analysis against established CNNs (GoogLeNet, ResNet18, DenseNet121, VGG) on four distinct datasets.

Main Results:

• The proposed IVGG networks demonstrate superior classification accuracies compared to VGG and other benchmark CNNs.
• IVGG networks exhibit a simpler architecture with a reduced number of parameters.
• The integration of the Inception module effectively enhances feature extraction for radar targets.
• Improved non-linearity strengthens the network's ability to discern subtle target differences.

Conclusions:

• IVGG networks represent a significant advancement in high-resolution radar target recognition.
• The proposed architecture offers a more efficient and accurate solution for classifying HRRP and SAR signals.
• Further research can explore broader applications of IVGG networks in radar signal processing.