YOLO-Extract: Improved YOLOv5 for Aircraft Object Detection in Remote Sensing Images

Abstract

The paper presents the YOLO-extract
algorithm, an enhancement of the YOLOv5 model
tailored for improving aircraft detection in remote
sensing images. Remote sensing targets pose
challenges due to their small and dense shapes
against complex backgrounds, leading to
insufficient detection accuracy and imprecise target
localization. The YOLO-extract algorithm
optimizes the model structure of YOLOv5,
incorporating features like a coordinate attention
mechanism and improved loss functions to address
these challenges. By focusing on enhancing the
ability to detect aircraft in remote sensing images,
the project contributes to advancements in satellite-
based object detection, particularly crucial in
applications such as airport monitoring and military
intelligence. Stakeholders including airport
authorities, military intelligence agencies, and
decision-makers in military operations benefit from
the improved detection capabilities, facilitating
better airport management, precise intelligence
analysis, and faster decision-making during
military actions. Furthermore, the paper suggests
exploring additional techniques such as
YOLOv5x6 and YOLOv8 to further enhance
performance. As an extension, the paper proposes
building a user-friendly front end using the Flask
framework for user testing with authentication,
enhancing the practical applicability and usability
of the proposed algorithms.