Related Experiment Video
Updated: Aug 14, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
PFE-Det: Progressive Feature Evolution for Small Object Detection in UAV Aerial Images
Aolin Fang1, Yongzi Zhang1, Xiaotong Dong2
1College of Mathematics and Computer Science, Guangdong Ocean University, Zhanjiang 524088, China.
Sensors (Basel, Switzerland)
|August 13, 2026
Summary
This study introduces the Progressive Feature Evolution Detector (PFE-Det) to improve small object detection in UAV aerial images by addressing feature degradation. PFE-Det enhances feature preservation and multi-scale context, significantly boosting small object detection precision.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Remote Sensing
Background:
- Object detection in UAV aerial imagery faces challenges with small objects, background noise, and feature degradation.
- Existing methods struggle with strided convolutions and feature overwriting, losing crucial details.
Purpose of the Study:
- To develop an advanced object detection method, PFE-Det, for enhanced small object recognition in UAV images.
- To overcome limitations in feature extraction pipelines that degrade structural information.
Main Methods:
- Proposed PFE-Det (Progressive Feature Evolution Detector) using a three-stage continuous optimization pathway.
- Implemented Feature Adaptive Enhancement Network (FAENet) for input-level detail preservation.
- Introduced Multi-Receptive-Field Adaptive Fusion Module (MFAM) and Multi-Path Gated State Space Modeling Block (MG-SSM Block) for feature reorganization and context modeling.
Main Results:
- Achieved an AP of 0.225 and APs of 0.134 on the VisDrone2019 dataset.
- Demonstrated a 13.5% relative improvement in small-object precision compared to baseline methods.
- Validated effectiveness on DIOR and UAVVaste datasets, confirming robust performance.
Conclusions:
- PFE-Det effectively addresses feature degradation issues in UAV small object detection.
- The proposed progressive feature evolution strategy significantly enhances detection accuracy.
- The method shows strong generalization capabilities across different UAV datasets.
