Motion feature extraction using magnocellular-inspired spiking neural networks for drone detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Neuroscience

Background:

• Traditional object detection struggles with small drones due to similar target and background features.
• Detecting small, low-altitude drones against complex backgrounds remains a significant challenge.

Purpose of the Study:

• To develop an improved drone detection system by leveraging bio-inspired motion processing.
• To enhance the accuracy of detecting small drones in challenging visual environments.

Main Methods:

• Proposed the Magno-Spiking Neural Network (MG-SNN), inspired by magnocellular motion processing.
• Developed a novel backpropagation method, Dynamic Threshold Multi-frame Spike Time Sequence (DT-MSTS).
• Created a dedicated dataset for training and validating the MG-SNN for drone detection.

Main Results:

• The MG-SNN effectively estimates motion saliency to identify potential moving targets.
• Integration of MG-SNN with existing object detection algorithms significantly boosts accuracy.
• The system acts as a plug-and-play module, enhancing detection of small flying targets.

Conclusions:

• The MG-SNN offers a significant improvement over conventional methods for small drone detection.
• The proposed method effectively combines motion and spatial features for superior detection accuracy.
• MG-SNN provides a cost-effective solution for detecting small drones in complex backgrounds.