Detection of low-altitude infrared small targets for UAVs using a density-based artificial bee colony algorithm

Area of Science:

• Computer Vision and Image Processing
• Artificial Intelligence and Machine Learning
• Aerospace Engineering and Remote Sensing

Background:

• Low-altitude UAVs struggle with detecting small targets in weak thermal signal conditions, limiting their operational effectiveness.
• Existing detection models lack the necessary accuracy and robustness for dynamic environments and subtle thermal signatures.
• The need for advanced algorithms to enhance infrared small target detection is critical for surveillance and monitoring applications.

Purpose of the Study:

• To develop an enhanced small target detection model for UAVs operating at low altitudes with weak thermal signals.
• To improve detection accuracy, robustness, and efficiency in challenging environmental conditions.
• To provide a technical solution for reliable infrared target detection in security, rescue, and monitoring missions.

Main Methods:

• Proposed an enhanced small target detection model integrating density-peak clustering and artificial bee colony optimization (DBABC algorithm).
• Implemented multi-stage infrared image preprocessing: Butterworth low-pass filtering, local background subtraction, and exponential high-pass filtering.
• Incorporated size change perception and environmental perturbation correction mechanisms for dynamic target adaptation.

Main Results:

• Achieved high detection accuracy: 91.66% on FLIR dataset and 90.38% on KAIST dataset, with a recall rate over 89.6%.
• Demonstrated significant signal-to-noise ratio gain (>23.19 dB) for long-distance detection (150-200 m).
• Maintained low detection delay (28-36 ms) and high resolution, outperforming advanced models, with proven robustness in adverse weather and low thermal contrast.

Conclusions:

• The proposed DBABC model significantly enhances small target detection accuracy and robustness for low-altitude UAVs in challenging infrared scenarios.
• The method offers a reliable technical pathway for critical applications like security patrols, disaster search and rescue, and border monitoring.
• This study advances the application of density-driven intelligent optimization in infrared small target detection, providing valuable theoretical support.