Proposed CORE-Net, a efficient dual-branch architecture for multispectral (RGB-IR) object detection that circumvents the heavy computational overhead of traditional cross-modal fusion paradigms. This framework demonstrates superior accuracy and robust performance in low-illumination environments, rendering it well-suited for deployment on resource-constrained edge devices.

Developed GOOD-Net, a lightweight and robust detection algorithm tailored for the challenges of small objects in drone imagery. This work establishes a new paradigm for efficient deep learning, validating its scalability and precision through extensive deployment on edge devices.

Proposed LCFF-Net, a computationally efficient algorithm that significantly improves tiny target detection performance in UAV imagery through improved cross-scale feature fusion. The model outperforms state-of-the-art baselines in both accuracy and inference speed, offering a robust solution for real-time aerial applications.