Research on Aerial Image Object Detection Methods Based on YOLOv7 with Different Attention Mechanisms

This study explores the effect of integrating attention mechanisms into the YOLOv7 architecture for aerial image object detection, aiming to enhance the model's capability to focus on critical regions and improve detection accuracy. Specifically, the Convolutional Block Attention Module (CBAM) and Efficient Channel Attention (ECA) module were embedded into YOLOv7 at layers 25 and 39, targeting stages where mid-level and high-level features are processed. These layers were strategically selected to capitalize on attention mechanisms for refining spatial and semantic feature representation. Despite these efforts, experimental results revealed that these modifications did not yield significant performance improvements. Including CBAM and ECA, we introduced additional computational overhead without enhancing detection accuracy relative to the original YOLOv7 model. This outcome indicates that the unique characteristics of aerial imagery—such as large, homogeneous backgrounds and relatively larger object scales—may limit the effectiveness of attention mechanisms in this context. These results align with broader findings that highlight the domain-specific nature of attention modules. Moving forward, alternative strategies for optimizing YOLOv7 for aerial imagery will be explored, focusing on balancing model accuracy and computational efficiency.