L-ENet: An Ultralightweight SAR Image Detection Network

Abstract

Synthetic aperture radar imagery plays a vital role in maritime vessel detection, despite facing challenges like complex marine environments, low-contrast target recognition, and similarities between ships and sea surface waves. In addition, the deployment of traditional large models on radar-equipped edge devices is hampered by their extensive computational demands. Addressing these challenges, this study presents an ultralightweight network lightweight enhanced network (L-ENet), improving upon YOLOv5n. The backbone network is switched to a more efficient ShufflenetV2, integrated with an improved attention mechanism, dual-core fusion attention mechanism, for effective interchannel information balancing. In the network's neck, the original C3 module has been replaced with the C3EGhost convolution. This convolution module integrates the Ghost and efficient channel attention mechanisms, aiming to mitigate potential accuracy loss during the lightweight process. Furthermore, this study introduces a novel multiscale fusion pathway and a Concat module with adaptive weights to better harmonize semantic and detail information. Furthermore, the trihead detection structure is revised to a dual-head structure omni-dimensional adaptive spatial feature fusion, using object detection convolution to allocate weights across different scales for efficient detection. Experimental results show that L-ENet has a computational cost of 0.6 M and a parameter count of 2.1 giga floating point operations per second. These figures are 65% and 50% lower than those of the original YOLOv5n while maintaining the same detection accuracy of 97.8%. Hence, L-ENet offers an efficient and viable solution for maritime target detection, showcasing commendable detection performance in comparison with advanced models.

Keywords

• Attention mechanisms
• lightweight network
• low-contrast target recognition
• synthetic aperture radar (SAR)