Detection of Inflatable Boats and People in Thermal Infrared with Deep Learning Methods.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Border Security Technology

Background:

• Illegal smuggling via small inflatable boats poses a significant financial threat to the European Union.
• Early detection is hampered by challenging environmental conditions and difficult terrain along border rivers.

Purpose of the Study:

• To develop and evaluate an automated system for detecting small inflatable boats and people in infrared thermal imagery.
• To assess the performance and processing time of various deep learning algorithms for this detection task.

Main Methods:

• Three deep learning algorithms (YOLOv2, YOLOv3, Faster R-CNN) were trained and tested.
• Six different feature extraction neural networks were utilized in conjunction with the algorithms.
• Data was collected across three seasonal acquisition campaigns under diverse weather conditions.

Main Results:

• Faster R-CNN combined with the ResNet101 feature extractor demonstrated the highest detection performance.
• While accurate, this optimal configuration demanded substantial processing time and powerful GPU resources.
• Performance and processing times varied across different algorithms and feature extractors.

Conclusions:

• Deep learning, particularly Faster R-CNN with ResNet101, shows promise for automated detection of smuggling activities in challenging environments.
• The trade-off between detection accuracy and computational cost needs consideration for practical deployment.
• Further research may focus on optimizing algorithms for real-time processing on less powerful hardware for enhanced border surveillance.