Real-time airplane detection using multi-dimensional attention and feature fusion.

Area of Science:

• Computer Vision
• Remote Sensing Technology
• Artificial Intelligence

Background:

• Airplane object detection in remote sensing images faces challenges like low resolution and background noise, leading to missed and misdetections.
• Existing methods struggle with accurately identifying small airplane objects amidst complex backgrounds.

Purpose of the Study:

• To develop a more accurate and efficient algorithm for airplane detection in remote sensing images.
• To address the limitations of current object detection models in handling low-resolution and noisy imagery.

Main Methods:

• Proposed the Accurate and Efficient YOLOv4-tiny (AE-YOLO) algorithm, incorporating a multi-dimensional channel and spatial attention module to filter background noise.
• Implemented a local cross-channel interaction strategy without dimensionality reduction to preserve local information.
• Utilized weighted two-way feature pyramid operations for enhanced feature fusion and channel correlation learning.
• Reconstructed the network using a lightweight convolution module to reduce parameters and computations.

Main Results:

• The AE-YOLO algorithm demonstrated improved detection precision for airplanes in remote sensing images.
• The proposed method is more lightweight and efficient compared to the original algorithm.
• Achieved a 7.76% higher detection accuracy than the baseline algorithm on the airplane dataset.
• The algorithm meets real-time detection requirements.

Conclusions:

• The AE-YOLO algorithm effectively overcomes challenges in remote sensing airplane detection, offering enhanced accuracy and efficiency.
• The integration of attention mechanisms and optimized feature fusion significantly improves detection performance.
• AE-YOLO provides a viable solution for real-time, high-accuracy airplane detection in challenging remote sensing scenarios.