Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HMPNet: A Feature Aggregation Architecture for Maritime Object Detection from a Shipborne Perspective

Yu Zhang, Fengyuan Liu, Juan Lyu, Yi Wei, Changdong Yu

TL;DR

The paper tackles the challenge of maritime object detection from shipborne perspectives by introducing Navigation12, a large, diverse dataset for 12 target categories, and presenting HMPNet, a lightweight architecture designed for robust, multi-scale feature aggregation. The architecture combines a hierarchical dynamic modulation HDM backbone, a matrix cascading poly-scale MCP neck, and a polymerization weight sharing PWS detector to achieve high accuracy with low parameter count. Empirical results show that HMPNet achieves state-of-the-art performance on Navigation12, delivering strong mAP while significantly reducing parameters and FLOPs compared to existing methods, enabling real-time inference. This work advances practical maritime perception by providing a dedicated dataset and a tailored, efficient detection framework suitable for shipborne navigation and safety applications.

Abstract

In the realm of intelligent maritime navigation, object detection from a shipborne perspective is paramount. Despite the criticality, the paucity of maritime-specific data impedes the deployment of sophisticated visual perception techniques, akin to those utilized in autonomous vehicular systems, within the maritime context. To bridge this gap, we introduce Navigation12, a novel dataset annotated for 12 object categories under diverse maritime environments and weather conditions. Based upon this dataset, we propose HMPNet, a lightweight architecture tailored for shipborne object detection. HMPNet incorporates a hierarchical dynamic modulation backbone to bolster feature aggregation and expression, complemented by a matrix cascading poly-scale neck and a polymerization weight sharing detector, facilitating efficient multi-scale feature aggregation. Empirical evaluations indicate that HMPNet surpasses current state-of-the-art methods in terms of both accuracy and computational efficiency, realizing a 3.3% improvement in mean Average Precision over YOLOv11n, the prevailing model, and reducing parameters by 23%.

HMPNet: A Feature Aggregation Architecture for Maritime Object Detection from a Shipborne Perspective

TL;DR

The paper tackles the challenge of maritime object detection from shipborne perspectives by introducing Navigation12, a large, diverse dataset for 12 target categories, and presenting HMPNet, a lightweight architecture designed for robust, multi-scale feature aggregation. The architecture combines a hierarchical dynamic modulation HDM backbone, a matrix cascading poly-scale MCP neck, and a polymerization weight sharing PWS detector to achieve high accuracy with low parameter count. Empirical results show that HMPNet achieves state-of-the-art performance on Navigation12, delivering strong mAP while significantly reducing parameters and FLOPs compared to existing methods, enabling real-time inference. This work advances practical maritime perception by providing a dedicated dataset and a tailored, efficient detection framework suitable for shipborne navigation and safety applications.

Abstract

In the realm of intelligent maritime navigation, object detection from a shipborne perspective is paramount. Despite the criticality, the paucity of maritime-specific data impedes the deployment of sophisticated visual perception techniques, akin to those utilized in autonomous vehicular systems, within the maritime context. To bridge this gap, we introduce Navigation12, a novel dataset annotated for 12 object categories under diverse maritime environments and weather conditions. Based upon this dataset, we propose HMPNet, a lightweight architecture tailored for shipborne object detection. HMPNet incorporates a hierarchical dynamic modulation backbone to bolster feature aggregation and expression, complemented by a matrix cascading poly-scale neck and a polymerization weight sharing detector, facilitating efficient multi-scale feature aggregation. Empirical evaluations indicate that HMPNet surpasses current state-of-the-art methods in terms of both accuracy and computational efficiency, realizing a 3.3% improvement in mean Average Precision over YOLOv11n, the prevailing model, and reducing parameters by 23%.
Paper Structure (11 sections, 1 equation, 5 figures, 2 tables)

This paper contains 11 sections, 1 equation, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. HDM Backbone
  4. MCP Neck
  5. PWS Detector
  6. Experiment
  7. Datasets
  8. Details and Evaluation Metrics
  9. Comparative Analysis
  10. Ablation Experiments
  11. Conclusion

Figures (5)

  • Figure 1: The architecture of the proposed HMPNet consists of three components, including (a) the HDM backbone, (b) the MCP neck, and (c) the PWS detector.
  • Figure 2: The structures of HGStem and HDFBlock. (a) HGStem, utilizes multi-scale convolution and pooling operations to extract initial features. (b) HDFBlock, performs hierarchical dynamic feature aggregation through dynamic convolutions and depthwise separable convolutions.
  • Figure 3: Structure of MCPC module.
  • Figure 4: Structure of PWS detector. The PWS detector reduces the number of parameters through shared convolution, incorporates GNDConv to enhance detail capture, and utilizes the Scale layer for adaptive adjustment of multi-scale features, ensuring detection accuracy and robustness.
  • Figure 5: Visualization Results of Different Detection Models show a comparison of our method with four state-of-the-art algorithms across three complex maritime scenarios—multi-scale multi-target, low-light at dusk, and wide depth of field—highlighting the superior performance of our method in these challenging conditions.