Real-Time Aircraft Target Detection Using Improved Yolov7-Tiny Network

Abstract

Detecting aircraft from remote sensing images has become increasingly important in applications such as military surveillance, airspace monitoring, airport management, and disaster response. However, identifying aircraft in satellite imagery is not straightforward. The objects are often small in size, surrounded by complex backgrounds, and affected by variations in lighting and resolution. Traditional object detection approaches and computationally heavy deep learning models either lack accuracy or fail to achieve real-time performance. Traditional computer vision techniques based on handcrafted features lack robustness and adaptability to such diverse conditions. Although deep learning-based object detection models have achieved remarkable success in natural image datasets, their direct application to remote sensing imagery does not always yield optimal performance, particularly for small object detection. Lightweight models designed for real-time inference often sacrifice detection accuracy to maintain speed. In this study, we propose an improved lightweight detection framework based on YOLOv7 tiny architecture for real-time aircraft detection in remote sensing images. The proposed method enhances multi-scale feature extraction and optimizes anchor box configurations to improve small-target detection performance while preserving computational efficiency. The model is implemented using Py-Torch and trained on annotated aircraft datasets. Experimental results demonstrate improved precision, recall, and mean Average Precision (map) compared to conventional lightweight detectors. The proposed system achieves a balance between detection accuracy and inference speed, making it suitable for practical real-world surveillance applications.