Bio-Inspired Feedback Visual Network for Robust Small-Target Motion Detection in Complex Environments.

Area of Science:

• Computer Science
• Artificial Intelligence
• Computer Vision

Background:

• Detecting and tracking small objects in complex, dynamic environments is a significant challenge for artificial intelligence (AI) vision systems.
• Limited spatial/textural features of small targets and misleading motion cues from dynamic backgrounds hinder reliable target discrimination.

Purpose of the Study:

• To propose an enhanced visual neural network model inspired by insect brains for improved small object detection and tracking.
• To address the limitations of current AI vision systems in distinguishing small targets from background noise and false motion signals.

Main Methods:

• Developed a visual neural network model incorporating a feedback mechanism for adaptive temporal response.
• Implemented a two-pathway architecture: motion detection and feedback attention.
• Integrated a global inhibition module to reduce false positives and improve overall detection.

Main Results:

• The proposed model achieved a detection rate of 0.81 in complex visual scenarios, outperforming compared models (below 0.59).
• Achieved superior Precision (0.0648) and F1-score (0.12) compared to existing methods (Precision < 0.0077, F1-score < 0.015).

Conclusions:

• The novel feedback-enhanced AI vision model demonstrates significant improvements in detecting and tracking small objects.
• The model's ability to adapt to temporal variations and filter false positives enhances robustness and accuracy in challenging environments.