Cross-modal edge-enhanced detector for UAV-based multispectral object detection.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Unmanned Aerial Vehicle (UAV)-based multispectral object detection is crucial for smart city traffic management and disaster response.
• Existing methods often overlook edge blurring in infrared imagery, complicating foreground-background distinction and object detection accuracy.

Purpose of the Study:

• To develop a novel cross-modal edge-enhanced detector to address challenges in UAV-based multispectral object detection.
• To improve the robustness and accuracy of object detection in adverse conditions by enhancing edge features.

Main Methods:

• Proposed an Edge Feature Enhancement Module using differential convolution to sharpen object edges in infrared images.
• Implemented a Multi-Scale Feature Fusion Module with dilated convolution for detecting objects of various sizes and adapting to resolution changes.
• Introduced a Cross-Modal Feature Fusion Module with a self-attention mechanism to effectively fuse complementary information from both visual and infrared spectra.

Main Results:

• The proposed CMEE-Det significantly enhances edge features, improving the distinction between objects and background.
• The model demonstrates superior performance in detecting objects of varying scales and adapting to UAV flight dynamics.
• Experimental results show that CMEE-Det outperforms existing methods on benchmark datasets like DroneVehicle.

Conclusions:

• The novel cross-modal edge-enhanced detector effectively addresses the limitations of existing methods in UAV-based multispectral object detection.
• The integration of edge enhancement and multi-modal fusion strategies leads to more robust and accurate object detection capabilities.
• This work offers a promising advancement for applications requiring reliable object detection from multispectral UAV imagery.