Current Projects

Objectives:

• Comprehensive Dataset Development: Generate a publicly available RF and communication waveform dataset comprising 56 distinct modulation classes, including both radar and communication waveforms, to address current limitations in existing datasets.

• Reproducible and Transparent Methodology: Provide detailed scripts and documentation openly via GitHub to enable straightforward reproduction, facilitating transparency and ease of use by the research community.

• Educational Resource: Serve as an educational resource for researchers and students to gain insights into waveform modulation techniques through clearly explained code and examples.

• Standardized Benchmarking: Offer a standardized, diverse benchmark dataset to support fair comparison and validation of emerging waveform classification methods in RF and communication signal processing research.

Objectives:

1. Multimodal Sensor Deployment:
   Develop, deploy, and validate advanced multi-sensor systems, including spectral-polarimetric imaging (across UV, VIS, NIR, SWIR, and LWIR bands), radar-based micro-Doppler sensing, and acoustic sensing arrays, to robustly detect and classify chemical and biological threats in subterranean environments.

2. Enhanced Sensor Fusion Techniques:
   Create and evaluate innovative sensor fusion strategies (data-level, feature-level, and decision-level fusion) to significantly improve the accuracy and reliability of threat detection, localization, and classification under complex and dynamic subterranean conditions.

3. Realistic Data Collection and Analysis:
   Generate comprehensive and unique multimodal datasets by conducting experiments in controlled laboratory settings and realistic field scenarios, using robotic platforms mimicking biological threat behaviors, to rigorously validate sensor performance and fusion algorithms.

4. Optimal Sensor Selection and Practical Guidelines:
   Investigate and provide recommendations on selecting optimal sensor combinations to maximize threat detection capabilities while balancing practical constraints such as power consumption, measurement complexity, and processing limitations for real-world deployment by first responders and military personnel.

Objectives:

• Develop a Computationally Efficient Model: Design a neural network architecture integrating parameterized 2D Gaussian filters to significantly reduce computational complexity for real-time ultrasound image classification.

• Enhance Real-time Classification Accuracy: Achieve improved or comparable accuracy to state-of-the-art CNN models (VGG-16, ResNet-50, DenseNet-121, AlexNet) while utilizing fewer computational resources.

• Validate Model Robustness: Conduct comparative experiments against widely used neural networks to rigorously assess performance, robustness, and efficiency of the proposed Gaussian-filter-based approach.

• Advance Practical Application: Demonstrate the suitability of learnable Gaussian filters as a practical, real-time medical imaging solution, enabling efficient deployment in clinical ultrasound diagnostics.