Adaptive condition-aware high-dimensional decoupling remote sensing image object detection algorithm.

Area of Science:

• Computer Science
• Remote Sensing
• Artificial Intelligence

Background:

• Remote Sensing Image Object Detection (RSIOD) is crucial but challenged by multi-scale objects, dense overlaps, and uneven data distribution.
• Existing methods often struggle with the scale, orientation, and shape variations inherent in remote sensing imagery.

Purpose of the Study:

• To propose a novel algorithm, YOLO-ACPHD, for improved RSIOD.
• To address the limitations of scale variation, object overlap, and computational efficiency in current RSIOD techniques.

Main Methods:

• The YOLO-ACPHD algorithm utilizes Adaptive Condition Awareness Technology (ACAT) for dynamic convolution kernel adjustment.
• High-Dimensional Decoupling Technology (HDDT) is employed to reduce computational complexity for large-scale remote sensing image processing.

Main Results:

• The YOLO-ACPHD model achieved a 0.99 F1 score, 1 Precision, 0.994 Precision-Recall, 1 Recall, and 99.36% mAP on the RSOD dataset.
• Demonstrated superior accuracy and robustness in detecting objects across various scales and densities.

Conclusions:

• The proposed YOLO-ACPHD algorithm significantly enhances the performance of RSIOD.
• The integration of ACAT and HDDT offers a robust and efficient solution for practical remote sensing applications.