Enhanced YOLOv8 model for accurate and real-time remote sensing target detection

• The highlights of our paper titled "Enhanced YOLOv8 Model for Accurate and Real-Time Remote Sensing Target Detection" are presented below:. • The PIYN model addresses common challenges in remote sensing image object detection, such as false positives, missed targets, and inadequate feature extraction for small-sized objects. • The integration of the EMA mechanism into the C2f module enhances the model's ability to focus on relevant regions, improving detection accuracy in complex backgrounds. • The Compact-PAN structure, introduced in the neck network, significantly reduces the computational load while maintaining high accuracy, making the model suitable for deployment on edge devices. • The Weighted Intersection over Union (WIoU) loss function improves the localization accuracy of detection boxes, enhancing the model's performance in real-world scenarios. • The study includes a comprehensive ablation experiment to validate the contribution of each component in the PIYN architecture, confirming the effectiveness of the proposed enhancements. • Comparative experiments with existing YOLO models demonstrate that PIYN achieves superior precision, recall, and mAP, while requiring less computational resources, highlighting its efficiency and practicality. • The research emphasizes the importance of data augmentation and cross-validation techniques in mitigating overfitting and ensuring robust model performance. • The PIYN model's capabilities are particularly relevant for applications in smart cities, such as real-time traffic monitoring, environmental surveillance, and infrastructure management, due to its enhanced accuracy and efficiency. Current remote sensing image object detection algorithms often struggle with false positives, missed targets, and suboptimal accuracy. To address these issues, we propose an improved YOLOv8 network (PIYN) solution achieved through targeted modifications to the YOLOv8 architecture. The backbone of YOLOv8 utilizes a Cross-Stage Partial (CSP) structure that includes two convolutions, called a faster C2f module. Firstly, we infuse the C2f module integrating an Efficient Multi-Scale Attention (EMA) mechanism, which enhances the module's ability to process information across various scales. Secondly, we introduce a Compact Path Aggregation Network (Compact-PAN) structure within the neck of the network, which reduces the computational complexity of the model. Finally, replacing the Complete Intersection over Union (CIoU) loss function with the Weighted Intersection over Union (WIoU) loss refines the model's detection accuracy. Additionally, we applied K-fold cross-validation on the dataset to mitigate overfitting. Experiments using the extensive Dataset for Object Detection in Aerial images (DOTA) and the Dataset for Object Recognition in Optical Remote Sensing Imagery (DIOR) reveal PIYN's effectiveness: there is a 2.43 % and 2.56 % increase in Mean Average Precision (mAP) over YOLOv8, respectively, alongside a 4.49 % reduction in GFLOPs. These results demonstrate PIYN's capability to enhance accuracy while maintaining efficiency and solidify its progressive and practical impact, particularly for smart city applications.