Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A drone detector with modified backbone and multiple pyramid featuremaps enhancement structure (MDDPE)

Chenhao Wu

TL;DR

This paper tackles the challenge of accurate, real-time drone detection from visual data, especially for small drones in complex backgrounds. It introduces MDDPE, a one-stage detector built on a modified D-LinkNet-based backbone, augmented by two feature-map enhancement modules (FMSE and FMRE) and a Tailored Improved Anchor Matching (TIAM) scheme to better detect drones across scales. Through ablations on a mixed, multi-source drone dataset and comparisons with state-of-the-art detectors, MDDPE delivers superior accuracy (notably AP50-95 improvements) while maintaining around 21 FPS on a high-end GPU, across Real World, Det-Fly, and MIDGARD benchmarks. The work demonstrates robust, real-world-ready drone detection and outlines directions for future improvements, including attention mechanisms and broader comparisons across detection scenarios.

Abstract

This work presents a drone detector with modified backbone and multiple pyramid feature maps enhancement structure (MDDPE). Novel feature maps improve modules that uses different levels of information to produce more robust and discriminatory features is proposed. These module includes the feature maps supplement function and the feature maps recombination enhancement function.To effectively handle the drone characteristics, auxiliary supervisions that are implemented in the early stages by employing tailored anchors designed are utilized. To further improve the modeling of real drone detection scenarios and initialization of the regressor, an updated anchor matching technique is introduced to match anchors and ground truth drone as closely as feasible. To show the proposed MDDPE's superiority over the most advanced detectors, extensive experiments are carried out using well-known drone detection benchmarks.

A drone detector with modified backbone and multiple pyramid featuremaps enhancement structure (MDDPE)

TL;DR

This paper tackles the challenge of accurate, real-time drone detection from visual data, especially for small drones in complex backgrounds. It introduces MDDPE, a one-stage detector built on a modified D-LinkNet-based backbone, augmented by two feature-map enhancement modules (FMSE and FMRE) and a Tailored Improved Anchor Matching (TIAM) scheme to better detect drones across scales. Through ablations on a mixed, multi-source drone dataset and comparisons with state-of-the-art detectors, MDDPE delivers superior accuracy (notably AP50-95 improvements) while maintaining around 21 FPS on a high-end GPU, across Real World, Det-Fly, and MIDGARD benchmarks. The work demonstrates robust, real-world-ready drone detection and outlines directions for future improvements, including attention mechanisms and broader comparisons across detection scenarios.

Abstract

This work presents a drone detector with modified backbone and multiple pyramid feature maps enhancement structure (MDDPE). Novel feature maps improve modules that uses different levels of information to produce more robust and discriminatory features is proposed. These module includes the feature maps supplement function and the feature maps recombination enhancement function.To effectively handle the drone characteristics, auxiliary supervisions that are implemented in the early stages by employing tailored anchors designed are utilized. To further improve the modeling of real drone detection scenarios and initialization of the regressor, an updated anchor matching technique is introduced to match anchors and ground truth drone as closely as feasible. To show the proposed MDDPE's superiority over the most advanced detectors, extensive experiments are carried out using well-known drone detection benchmarks.
Paper Structure (22 sections, 8 equations, 10 figures, 4 tables)

This paper contains 22 sections, 8 equations, 10 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. The problem
  4. The solution
  5. Pipeline of model
  6. Backbone Selection and Modification
  7. Feature maps supplement Function
  8. Feature maps recombination enhancement
  9. Tailored prior anchor design and anchors matching portfolio
  10. Implementation Details
  11. Drone dataset construction
  12. Evaluation
  13. Backbones and input sizes experiments
  14. Components Analysis on MDDPE
  15. Feature Maps Supplement Function
  16. ...and 7 more sections

Figures (10)

  • Figure 1: The statistics about yearly drone-related injuries published by US emergency departments.
  • Figure 2: The framework of MDDPE uses a modified D-Linknet Backbone in yellow frame, Feature supplement Function to provide supplement feature information in red frame, the green frame depicts the feature maps function and its enhancement methodology and legend in black solid cable frame.
  • Figure 3: The architecture of D-linkNet(left) and the modified backbone based on D-LinkNet(right) i.e The A, B, C in left represent encoder, centre connection and decoder respectively.
  • Figure 4: The map of recall on original matrix after a serial dilated convolution with dilation setting in [1,2,4](left) and [1,2,3](right)
  • Figure 5: The architecture of a up to bottom model in Feature maps recombination enhancement
  • ...and 5 more figures