LDMNet: Enhancing the Segmentation Capabilities of Unmanned Surface Vehicles in Complex Waterway Scenarios DOI Creative Commons

Tongyang Dai,

Huiyu Xiang,

Chongjie Leng

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(17), P. 7706 - 7706

Published: Aug. 31, 2024

Semantic segmentation-based Complex Waterway Scene Understanding has shown great promise in the environmental perception of Unmanned Surface Vehicles. Existing methods struggle with estimating edges obstacles under conditions blurred water surfaces. To address this, we propose Lightweight Dual-branch Mamba Network (LDMNet), which includes a CNN-based Deep for extracting image features and Mamba-based fusion module aggregating integrating global information. Specifically, improve structure by incorporating multiple Atrous branches local fusion; design Convolution-based Recombine Attention Module, serves as gate activation condition Mamba-2 to enhance feature interaction information from both spatial channel dimensions. Moreover, tackle directional sensitivity serialization impact State Space Model’s forgetting strategy on non-causal data modeling, introduce Hilbert curve scanning mechanism achieve multi-scale serialization. By stacking sequences, alleviate bias towards sequence data. LDMNet integrates Network, Attention, blocks, effectively capturing long-range dependencies context images. The experimental results four benchmarks show that proposed significantly improves obstacle edge segmentation performance outperforms existing across various metrics.

Language: Английский

Remote sensing image interpretation of geological lithology via a sensitive feature self-aggregation deep fusion network DOI Creative Commons
康利 千賀, Jie Dong, Haozheng Ma

et al.

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2025, Volume and Issue: 137, P. 104384 - 104384

Published: Feb. 26, 2025

Language: Английский

Citations

1
Research progress in water quality prediction based on deep learning technology: a review DOI
Wenhao Li,

Yin Zhao,

Yining Zhu

et al.

Environmental Science and Pollution Research, Journal Year: 2024, Volume and Issue: 31(18), P. 26415 - 26431

Published: March 27, 2024

Language: Английский

Citations

5
BikeshareGAN: Predicting Dockless Bike-Sharing Demand Based on Satellite Image DOI
Yalei Zhu, Yuankai Wang, Junxuan Li

et al.

Published: Jan. 1, 2025

Language: Английский

Citations

0
BikeshareGAN: Predicting dockless bike-sharing demand based on satellite image DOI Creative Commons
Yalei Zhu, Yuankai Wang, Junxuan Li

et al.

Journal of Transport Geography, Journal Year: 2025, Volume and Issue: 126, P. 104245 - 104245

Published: April 28, 2025

Language: Английский

Citations

0
Review of Recent Advances in Remote Sensing and Machine Learning Methods for Lake Water Quality Management DOI Creative Commons
Ying Deng, Yue Zhang,

Daiwei Pan

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(22), P. 4196 - 4196

Published: Nov. 11, 2024

This review examines the integration of remote sensing technologies and machine learning models for efficient monitoring management lake water quality. It critically evaluates performance various satellite platforms, including Landsat, Sentinel-2, MODIS, RapidEye, Hyperion, in assessing key quality parameters chlorophyll-a (Chl-a), turbidity, colored dissolved organic matter (CDOM). highlights specific advantages each platform, considering factors like spatial temporal resolution, spectral coverage, suitability these platforms different sizes characteristics. In addition to this paper explores application a wide range models, from traditional linear tree-based methods more advanced deep techniques convolutional neural networks (CNNs), recurrent (RNNs), generative adversarial (GANs). These are analyzed their ability handle complexities inherent data, high dimensionality, non-linear relationships, multispectral hyperspectral data. also discusses effectiveness predicting parameters, offering insights into most appropriate model–satellite combinations scenarios. Moreover, identifies challenges associated with data quality, model interpretability, integrating imagery models. emphasizes need advancements fusion techniques, improved generalizability, developing robust frameworks multi-source concludes by targeted recommendations future research, highlighting potential interdisciplinary collaborations enhance sustainable management.

Language: Английский

Citations

1
Spatio-temporal data prediction of multiple air pollutants in multi-cities based on 4D digraph convolutional neural network DOI Creative Commons
Li Wang, Qianhui Tang, Xiaoyi Wang

et al.

PLoS ONE, Journal Year: 2023, Volume and Issue: 18(12), P. e0287781 - e0287781

Published: Dec. 22, 2023

In response to the problem that current multi-city multi-pollutant prediction methods based on one-dimensional undirected graph neural network models cannot accurately reflect two-dimensional spatial correlations and directedness, this study proposes a four-dimensional directed model can capture information node correlation related multiple factors, as well extract temporal at different times. Firstly, A GCN with in space was established geographical location of city. Secondly, Spectral decomposition tensor operations were then applied obtain Fourier coefficients basis. Thirdly, filter further improved optimized. Finally, an LSTM architecture introduced construct GCN-LSTM for synchronous extraction spatio-temporal atmospheric pollutant concentrations. The uses 2020 six-parameter data Taihu Lake city cluster applies canonical analysis confirm data's temporal, spatial, multi-factor correlations. Through experimentation, it is verified proposed 4D-DGCN-LSTM achieves MAE reduction 1.12%, 4.91%, 5.62%, 11.67% compared 4D-DGCN, GCN-LSTM, GCN, models, respectively, indicating good performance predicting types pollutants various cities.

Language: Английский

Citations

2
LDMNet: Enhancing the Segmentation Capabilities of Unmanned Surface Vehicles in Complex Waterway Scenarios DOI Creative Commons

Tongyang Dai,

Huiyu Xiang,

Chongjie Leng

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(17), P. 7706 - 7706

Published: Aug. 31, 2024

Semantic segmentation-based Complex Waterway Scene Understanding has shown great promise in the environmental perception of Unmanned Surface Vehicles. Existing methods struggle with estimating edges obstacles under conditions blurred water surfaces. To address this, we propose Lightweight Dual-branch Mamba Network (LDMNet), which includes a CNN-based Deep for extracting image features and Mamba-based fusion module aggregating integrating global information. Specifically, improve structure by incorporating multiple Atrous branches local fusion; design Convolution-based Recombine Attention Module, serves as gate activation condition Mamba-2 to enhance feature interaction information from both spatial channel dimensions. Moreover, tackle directional sensitivity serialization impact State Space Model’s forgetting strategy on non-causal data modeling, introduce Hilbert curve scanning mechanism achieve multi-scale serialization. By stacking sequences, alleviate bias towards sequence data. LDMNet integrates Network, Attention, blocks, effectively capturing long-range dependencies context images. The experimental results four benchmarks show that proposed significantly improves obstacle edge segmentation performance outperforms existing across various metrics.

Language: Английский

Citations

0