An Enhanced Lightweight Convolutional Neural Network for Ship Detection in Maritime Surveillance System

Abstract

With the extensive application of artificial intelligence, ship detection from optical satellite remote sensing images using deep learning technology can significantly improve detection accuracy. However, the existing methods usually have complex models and huge computations, which makes them difficult to deploy on resource-constrained devices, such as satellites. To solve this problem, this article proposes an enhanced lightweight ship detection model called ShipDetectionNet to replace the standard convolution with improved convolution units. The improved convolution unit is implemented by applying depthwise separable convolution to replace standard convolution and further using the pointwise group convolution to replace the point convolution in depthwise separable convolution. In addition, the attention mechanism is incorporated into the convolution unit to ensure detection accuracy. Compared to the latest YOLOv5s, our model has a comparable performance in mean average precision, while the number of parameters and the model size are reduced by 14.18% and 13.14%, respectively. Compared to five different lightweight detection models, the proposed ShipDetectionNet is more competent for ship detection tasks. In addition, the ShipDetectionNet is evaluated on four challenging scenarios, demonstrating its generalizability and effectiveness.

Keywords

• Attention mechanism
• depthwise separable convolution (DS-Conv)
• maritime surveillance system
• optical satellite remote sensing image
• pointwise group convolution (G-Conv)
• ship detection