FEMA-Net: A Fast Multi-Spatial Attention Mechanism for Ship Detection in Remote Sensing Images
Zhang Liu, Ying Wang, Jianbo Xu, Yulong Wang, Changxu Wan, Tiantian Zhang · 2023
The advancement of satellite technology enables the swift detection of ships in vast sea areas, which is crucial for effectively obtaining information. However, remote sensing images of ships exhibit features such as considerable width, reduced size, and significant shape variations. These attributes pose challenges for visual observation, and the complexity of backgrounds and weather interference exacerbates this challenge. To solve these problems, the present study FEMA-Net, a fresh ship detection model using remote sensing. Firstly, we enhance the C2f module by introducing the C2f-FNB module, optimizing its applicability on platforms with limited computational resources. Secondly, to alleviate the issue of overfitting, this paper introduces the EMA attention mechanism module. This addition directs the model's attention towards smaller ship targets, enhancing spatial feature capabilities, and reducing emphasis on larger backgrounds. Experimental results on a processed remote sensing ship detection dataset show significant improvements in both speed and accuracy compared to popular models.