Robust Ship Detection in Infrared Images through Multiscale Feature Extraction and Lightweight CNN

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Current ship detection technology in remote sensing images lacks sophistication.
• Existing methods struggle with multi-scene, multi-resolution, and multi-type target ships.
• Detection results often do not meet practical application requirements.

Purpose of the Study:

• To propose an improved ship detection method.
• To address limitations in handling diverse ship targets and complex environments.
• To enhance the accuracy and efficiency of ship detection in remote sensing.

Main Methods:

• Multiscale feature extraction for candidate region identification.
• Multiple feature fusion including Fourier global spectrum and lightweight CNN for classification.
• Cascade classifier training and improved non-maximum suppression for target confirmation.

Main Results:

• The proposed method significantly outperforms existing alternatives.
• Model size is reduced by up to 2.17 times.
• Detection performance improved by up to 5.5% in multi-interference scenarios.
• Robustness verified with F-measure score and true-false-positive rate increases (up to 5.8% and 4.7%), and mean error rate decrease (up to 38.2%).

Conclusions:

• The developed ship detection method is effective and robust.
• It offers superior performance compared to current approaches.
• The method provides a viable solution for practical, differentiated ship detection needs in remote sensing.