Sensing systems pervade modern life. Radar data feeds blind spot detection in cars and flight control in autonomous aircraft. Thermal cameras monitor agricultural crops and hostile missiles. Ultrasound diagnoses medical issues in humans and structural issues in bridges. Throughout private, commercial, industrial, and military sectors, sensors are increasingly relied on for critical—and dangerous—applications.

As reliance on these technologies grows, so too does the complexity of the environments in which they operate. Sensing systems are increasingly deployed in challenging conditions characterized by high levels of noise, rapid environmental changes, dynamic objects, and even hostile action. In critical applications, the consequences of failure can be severe, underscoring the need for robust sensing solutions.

This dissertation focuses on developing computational methods to tackle these critical challenges in modern sensing systems. By combining learning-based techniques with a physical understanding of sensing modalities, I aim to enhance the robustness, adaptability, and performance of these systems in complex and hostile environments. Specifically, we present three key contributions to: (1) design adaptive low probability of detection radar waveforms, (2) improve speckle denoising of dynamic scenes, and (3) advance high-speed imaging through turbulence-induced distortions.