Accurate detection of multiple small targets in a wide field of view based on the compound-eye imaging system

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Compound-eye imaging systems offer wide field of view (FOV) and high sensitivity to moving targets.
• These characteristics are advantageous for unmanned reconnaissance, large-scale monitoring, and dynamic object detection.
• Existing methods struggle with accurate small object identification in complex compound-eye imagery.

Purpose of the Study:

• To develop an innovative wide-FOV small object detection method using compound-eye imaging.
• To enhance the accuracy and reliability of detecting multiple categories of small targets.
• To establish a robust dataset and fusion strategy for compound-eye image analysis.

Main Methods:

• A convolutional attention super-resolution fusion network (CASFNet) was designed for super-resolution upsampling and adaptive feature fusion.
• A novel compound-eye sub-image (CESI) dataset was created, leveraging FOV-overlap for hardware-level data enhancement.
• A confidence-weighted fusion strategy was implemented, optimizing confidence scores using system-specific imaging parameters.

Main Results:

• The CASFNet achieved high performance in multi-category small object detection.
• Experimental validation showed a precision of 96.2% and a mean average precision (mAP) of 94.2%.
• The proposed method significantly enhanced the overall reliability of object detection in compound-eye imaging systems.

Conclusions:

• The developed compound-eye imaging method offers a significant advancement for small object detection.
• This technology has broad applications in wide-area surveillance and intelligent transportation systems.
• The combination of CASFNet, CESI dataset, and confidence-weighted fusion provides a robust solution for complex imaging scenarios.