Real-Time Multi-Scene Visibility Enhancement for Promoting Navigational Safety of Vessels Under Complex Weather Conditions

TL;DR

This work tackles the problem of degraded visible-light imagery for maritime navigation under complex weather by introducing ERANet, a lightweight, real-time network that performs multi-scene visibility enhancement. It combines Kirsch-gradient based edge reparameterization (KRM) with channel and spatial attention (CAM and SAM) to recover edge-preserving, color-natural images efficiently, merging multi-branch operations into a single convolution at inference. The method is trained with a hybrid loss that blends multi-scale structural similarity, L1 fidelity, and total-variation regularization, achieving strong performance on standard dehazing/deraining/low-light benchmarks and IWTS datasets, while also improving downstream YOLOv7 detection and DeepLabv3+ segmentation. Practically, ERANet delivers robust image restoration at over 40 frames per second on 1080p inputs with a small footprint, making it well-suited for onboard IWTS deployments and real-time navigational safety support.

Abstract

The visible-light camera, which is capable of environment perception and navigation assistance, has emerged as an essential imaging sensor for marine surface vessels in intelligent waterborne transportation systems (IWTS). However, the visual imaging quality inevitably suffers from several kinds of degradations (e.g., limited visibility, low contrast, color distortion, etc.) under complex weather conditions (e.g., haze, rain, and low-lightness). The degraded visual information will accordingly result in inaccurate environment perception and delayed operations for navigational risk. To promote the navigational safety of vessels, many computational methods have been presented to perform visual quality enhancement under poor weather conditions. However, most of these methods are essentially specific-purpose implementation strategies, only available for one specific weather type. To overcome this limitation, we propose to develop a general-purpose multi-scene visibility enhancement method, i.e., edge reparameterization- and attention-guided neural network (ERANet), to adaptively restore the degraded images captured under different weather conditions. In particular, our ERANet simultaneously exploits the channel attention, spatial attention, and reparameterization technology to enhance the visual quality while maintaining low computational cost. Extensive experiments conducted on standard and IWTS-related datasets have demonstrated that our ERANet could outperform several representative visibility enhancement methods in terms of both imaging quality and computational efficiency. The superior performance of IWTS-related object detection and scene segmentation could also be steadily obtained after ERANet-based visibility enhancement under complex weather conditions.

Figures (14)

• Figure 1: The workflow of ERANet-based real-time multi-scene low-visibility scene recovery for intelligent waterborne transportation systems (IWTS).
• Figure 2: The flowchart of our edge reparameterization- and attention-guided network (ERANet) for multi-scene visibility enhancement (i.e., hazy, rainy, and low-light). The applications of channel attention, spatial attention, and Kirsch eight-directions-guided edge parameterization are able to balance the low-visibility scene restoration and computational cost.
• Figure 3: The pipeline of Kirsch-guided reparameterization module (KRM). In the training stage, the KRM employs multiple branches, which can be merged into one normal convolutional layer in the inference stage.
• Figure 4: The convergence analysis under different degradation scenarios.
• Figure 5: Visual comparisons of dehazing results from Seaships shao2018seaships and SMD prasad2017video. From left to right: (a) hazy images, restored images, respectively, yielded by (b) SDD hao2020low, (c) ROP+ liu2022rank, (d) AirNet li2022all, (e) MIRNetv2 zamir2022learning, (f) TransWeather valanarasu2022transweather, (g) WeatherDiff ozdenizci2023restoring, (h) ERANet, and (i) Ground Truth.