Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

YOLO-FEDER FusionNet: A Novel Deep Learning Architecture for Drone Detection

Tamara R. Lenhard, Andreas Weinmann, Stefan Jäger, Tobias Koch

TL;DR

This work introduces a novel deep learning architecture called YOLO-FEDER FusionNet, which combines generic object detection methods with the specialized strength of camouflage object detection techniques to enhance drone detection capabilities.

Abstract

Predominant methods for image-based drone detection frequently rely on employing generic object detection algorithms like YOLOv5. While proficient in identifying drones against homogeneous backgrounds, these algorithms often struggle in complex, highly textured environments. In such scenarios, drones seamlessly integrate into the background, creating camouflage effects that adversely affect the detection quality. To address this issue, we introduce a novel deep learning architecture called YOLO-FEDER FusionNet. Unlike conventional approaches, YOLO-FEDER FusionNet combines generic object detection methods with the specialized strength of camouflage object detection techniques to enhance drone detection capabilities. Comprehensive evaluations of YOLO-FEDER FusionNet show the efficiency of the proposed model and demonstrate substantial improvements in both reducing missed detections and false alarms.

YOLO-FEDER FusionNet: A Novel Deep Learning Architecture for Drone Detection

TL;DR

This work introduces a novel deep learning architecture called YOLO-FEDER FusionNet, which combines generic object detection methods with the specialized strength of camouflage object detection techniques to enhance drone detection capabilities.

Abstract

Predominant methods for image-based drone detection frequently rely on employing generic object detection algorithms like YOLOv5. While proficient in identifying drones against homogeneous backgrounds, these algorithms often struggle in complex, highly textured environments. In such scenarios, drones seamlessly integrate into the background, creating camouflage effects that adversely affect the detection quality. To address this issue, we introduce a novel deep learning architecture called YOLO-FEDER FusionNet. Unlike conventional approaches, YOLO-FEDER FusionNet combines generic object detection methods with the specialized strength of camouflage object detection techniques to enhance drone detection capabilities. Comprehensive evaluations of YOLO-FEDER FusionNet show the efficiency of the proposed model and demonstrate substantial improvements in both reducing missed detections and false alarms.
Paper Structure (12 sections, 4 figures, 6 tables)

This paper contains 12 sections, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Framework
  4. Experimental Setup
  5. Datasets
  6. Evaluation Metrics
  7. Implementation Details
  8. Results
  9. Performance on Real-World Data
  10. Labeling Bias
  11. Drone Detection in an Alarm Scenario
  12. Conclusion

Figures (4)

  • Figure 1: Visual comparison between YOLO-FEDER FusionNet (red bounding boxes) and YOLOv5l (blue bounding box), showcasing every fifth image frame. YOLO-FEDER FusionNet consistently detects the drone across all six frames, whereas YOLOv5l only identifies it in the last one.
  • Figure 2: Overview of YOLO-FEDER FusionNet. Key components fusing and processing information from both backbones are highlighted in red. Layers are abbreviated as follows: CFE (camouflage feature encoder), CAM (channel attention module), CBS (convolution + batch normalization + SiLU activation), CBAM (convolutional block attention module), C3 (CSP bottleneck with three convolutional layers), SED (segmentation-oriented edge-assisted decoder), SPPF (spatial pyramid pooling fusion). The visualization is inspired by the illustration of the YOLOv5l architecture in Ultralytics.
  • Figure 3: Integration of CBAM into the bottleneck of the C3 module, located by default in the neck and head of YOLOv5l. Modified parts are highlighted in red.
  • Figure 4: Visual comparison of the manually labeled GT bounding boxes (blue) and the bounding boxes predicted by YOLO-FEDER FusionNet (red). While the GT boxes provide a more generous encapsulation of the drone, the predicted bounding boxes demonstrate a superior level of accuracy.