SP-Y OLOX: A Lightweight Ship Target Detection Algorithm Based on Remote Sensing Images

Gang Li,Xuanhong Wang,Dou Zhang,Ziyang Wu,Yun Xiao

DOI: https://doi.org/10.23919/ccc63176.2024.10662256

2024-01-01

Abstract:The detection of ships is a crucial and formidable task in remote sensing applications, with the accurate identification of ship targets playing a pivotal role in the advancement of maritime military forces and ensuring the safety of maritime navigation. However, the current ship detection model faces challenges due to its extensive parameter count, substantial computational requirements, and deployment limitations in resource-constrained environmental equipment. Henceforth, we propose a novel and efficient lightweight object detection algorithm, SP-Y OLOX. Firstly, a novel lightweight feature extraction module, PSCLay, is devised to ensure the efficacy of the feature extraction network while simultaneously reducing the parameter count and computational burden of the backbone network. Secondly, the FOCUS layer in the backbone network head incorporates a PSA attention mechanism to enhance spatial location information retention and capture features at various scales more effectively. Additionally, depth separable convolution is introduced in the detection header to mitigate model complexity. The training and testing were conducted on the remote sensing datasets DIOR and NWPU VHR-10. The experimental results demonstrate that our model, SP-Y OLOX, has a size of 3.21 M and a computational weight of 3.67 G. In comparison to the original Y OLOX baseline model, although there is a reduction in accuracy by $1.8 \%$, we achieve a decrease in parameter count by $36.3 \%$ and computational workload by $43.1 \%$.