Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TSOM: Small Object Motion Detection Neural Network Inspired by Avian Visual Circuit

Pignge Hu, Xiaoteng Zhang, Mengmeng Li, Yingjie Zhu, Li Shi

TL;DR

TSOM presents a biologically interpretable neural network for small object motion detection inspired by the avian Retina‑OT‑Rt circuit. It models the four-layer pathway and implements a two‑stage Rt integration to detect tiny moving targets in cluttered aerial scenes. The method demonstrates biological consistency with pigeon OT neurons and achieves superior detection accuracy on synthetic BEVS and real‑world RIST datasets compared with classical unsupervised methods and apg‑STMD in most scenarios. The work highlights the value of bio‑inspired spatiotemporal processing for robust small‑object tracking in remote sensing and UAV applications.

Abstract

Detecting small moving objects in complex backgrounds from an overhead perspective is a highly challenging task for machine vision systems. As an inspiration from nature, the avian visual system is capable of processing motion information in various complex aerial scenes, and its Retina-OT-Rt visual circuit is highly sensitive to capturing the motion information of small objects from high altitudes. However, more needs to be done on small object motion detection algorithms based on the avian visual system. In this paper, we conducted mathematical modeling based on extensive studies of the biological mechanisms of the Retina-OT-Rt visual circuit. Based on this, we proposed a novel tectum small object motion detection neural network (TSOM). The neural network includes the retina, SGC dendritic, SGC Soma, and Rt layers, each layer corresponding to neurons in the visual pathway. The Retina layer is responsible for accurately projecting input content, the SGC dendritic layer perceives and encodes spatial-temporal information, the SGC Soma layer computes complex motion information and extracts small objects, and the Rt layer integrates and decodes motion information from multiple directions to determine the position of small objects. Extensive experiments on pigeon neurophysiological experiments and image sequence data showed that the TSOM is biologically interpretable and effective in extracting reliable small object motion features from complex high-altitude backgrounds.

TSOM: Small Object Motion Detection Neural Network Inspired by Avian Visual Circuit

TL;DR

TSOM presents a biologically interpretable neural network for small object motion detection inspired by the avian Retina‑OT‑Rt circuit. It models the four-layer pathway and implements a two‑stage Rt integration to detect tiny moving targets in cluttered aerial scenes. The method demonstrates biological consistency with pigeon OT neurons and achieves superior detection accuracy on synthetic BEVS and real‑world RIST datasets compared with classical unsupervised methods and apg‑STMD in most scenarios. The work highlights the value of bio‑inspired spatiotemporal processing for robust small‑object tracking in remote sensing and UAV applications.

Abstract

Detecting small moving objects in complex backgrounds from an overhead perspective is a highly challenging task for machine vision systems. As an inspiration from nature, the avian visual system is capable of processing motion information in various complex aerial scenes, and its Retina-OT-Rt visual circuit is highly sensitive to capturing the motion information of small objects from high altitudes. However, more needs to be done on small object motion detection algorithms based on the avian visual system. In this paper, we conducted mathematical modeling based on extensive studies of the biological mechanisms of the Retina-OT-Rt visual circuit. Based on this, we proposed a novel tectum small object motion detection neural network (TSOM). The neural network includes the retina, SGC dendritic, SGC Soma, and Rt layers, each layer corresponding to neurons in the visual pathway. The Retina layer is responsible for accurately projecting input content, the SGC dendritic layer perceives and encodes spatial-temporal information, the SGC Soma layer computes complex motion information and extracts small objects, and the Rt layer integrates and decodes motion information from multiple directions to determine the position of small objects. Extensive experiments on pigeon neurophysiological experiments and image sequence data showed that the TSOM is biologically interpretable and effective in extracting reliable small object motion features from complex high-altitude backgrounds.
Paper Structure (34 sections, 1 theorem, 25 equations, 14 figures, 2 tables)

This paper contains 34 sections, 1 theorem, 25 equations, 14 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Related Works
  4. Methodology
  5. Mathematical Description of Retina-OT-Rt Circuit
  6. Spatiotemporal Dynamic of SGC
  7. Two-Stage Motion Information Integration of Rt
  8. TSOM Neural Network Model for Small Object Motion Detection
  9. Retina Layer
  10. SGC Dendrite Layer
  11. SGC Soma Layer
  12. Rt Layer
  13. Experiment and Results
  14. Datasets and Metrics
  15. Neuroelectrophysiological data
  16. ...and 19 more sections

Key Result

Proposition 1

For an output Rt neuron site, let $E_{2}$ be the activation probability of the two-stage framework and $E_{1}$ be that of the one-stage framework, then $E_{1}\leq E_{2}$.

Figures (14)

  • Figure 1: Neural pathway and model architecture of RGC-SGC-Rt neural pathway. (a)SGC-Rt two-stage connection graph. (b)RGC-SGC-Rt neural connection graph. (c)RGC-SGC-Rt neural network model.
  • Figure 2: Algorithm flow chart.
  • Figure 3: Diagram of the scale selection kernel function. (a) Schematic representation of the receptive field of a neuron; (b) the scale selection kernel function.
  • Figure 4: Examples from the datasets of (a) BEV and (b) RIST.
  • Figure 5: Signal processing validation experimental paradigm and results. (a) Input image $I\left ( u,v,t \right )$ with a resolution of 512×512. The background moves at a velocity of $V_{b}$ along a fixed direction, while the object moves at a velocity of $V_{a}$ along the direction $\theta=0$. The object's initial position is located at coordinates $\left ( 282,102 \right )$. (b) Results of directional selectivity validation. (c)-(j) The output of each module of the TSOM.
  • ...and 9 more figures

Theorems & Definitions (1)

  • Proposition 1