[Paper Review] A Feedback Neural Network for Small Target Motion Detection in Cluttered Backgrounds

Hongxin Wang, Jigen Peng+1 more|arXiv (Cornell University)|May 1, 2018

Neurobiology and Insect Physiology Research12 references3 citations

TL;DR

This paper proposes a feedback neural network (feedback ESTMD) that enhances small target motion detection in cluttered backgrounds by incorporating a temporally delayed feedback loop from the output to the medulla layer. The feedback suppresses responses to background noise, significantly improving detection performance over the baseline ESTMD model, especially when there is a velocity difference between the target and background.

ABSTRACT

Small target motion detection is critical for insects to search for and track mates or prey which always appear as small dim speckles in the visual field. A class of specific neurons, called small target motion detectors (STMDs), has been characterized by exquisite sensitivity for small target motion. Understanding and analyzing visual pathway of STMD neurons are beneficial to design artificial visual systems for small target motion detection. Feedback loops have been widely identified in visual neural circuits and play an important role in target detection. However, if there exists a feedback loop in the STMD visual pathway or if a feedback loop could significantly improve the detection performance of STMD neurons, is unclear. In this paper, we propose a feedback neural network for small target motion detection against naturally cluttered backgrounds. In order to form a feedback loop, model output is temporally delayed and relayed to previous neural layer as feedback signal. Extensive experiments showed that the significant improvement of the proposed feedback neural network over the existing STMD-based models for small target motion detection.

Motivation & Objective

• To investigate whether a feedback loop in the STMD neural pathway can improve small target motion detection performance.
• To address the challenge of detecting small, dim moving targets in naturally cluttered backgrounds where background noise mimics target features.
• To enhance the performance of existing STMD-based models by integrating a feedback mechanism that reduces false positives from background noise.
• To validate the effectiveness of feedback in improving detection rates under varying target and background motion conditions.

Proposed method

• The model extends the existing ESTMD architecture with a feedback loop that delays the output signal and feeds it back to the medulla layer.
• The feedback signal is added to the medulla layer's output to suppress responses to background noise while preserving target motion signals.
• The feedback mechanism is implemented using a temporal delay block to align the feedback signal with the current input stream.
• The model processes input image sequences through four layers: retina (Gaussian filtering), lamina (motion energy computation), medulla (feedback integration), and lobula (final output).
• The feedback loop is designed to weaken responses to static or slowly moving background features that resemble small targets.
• Performance is evaluated using ROC curves and detection rate at fixed false alarm rates across varying parameters such as target luminance, size, velocity, and background motion.

Experimental results

Research questions

• RQ1 Does incorporating a feedback loop into the STMD neural pathway significantly improve small target motion detection performance?
• RQ2 How does the feedback mechanism affect detection performance under varying target and background velocities?
• RQ3 Can the feedback network effectively suppress background noise while maintaining sensitivity to small moving targets?
• RQ4 How does the performance of the feedback ESTMD compare to the baseline ESTMD across different target and background motion parameters?

Key findings

• The feedback ESTMD model achieves significantly higher detection rates than the baseline ESTMD model at a fixed false alarm rate of 10.
• Under varying target luminance and size, feedback ESTMD consistently outperforms ESTMD, with detection rates substantially higher in low-contrast and small-target conditions.
• When target velocity exceeds background velocity (e.g., 500 vs. 250 px/s), feedback ESTMD shows markedly improved detection performance, especially in opposing motion scenarios.
• In same-direction motion, feedback ESTMD maintains superior performance when target velocity is higher than background velocity, but performance difference diminishes when background moves faster.
• In complex, naturally cluttered backgrounds (e.g., moving foliage), feedback ESTMD achieves higher detection rates than ESTMD, as confirmed by ROC curve comparisons.
• The feedback mechanism effectively reduces false positives from background noise, particularly in dynamic scenes with high texture similarity to small targets.