Abstract:Influenced by deep learning, extracting roads from high-resolution remote sensing images has attracted extensive attention. However, most previous works have focused on fully supervised models relying on large amounts of annotated data and have not considered the characteristics of narrow and elongated roads. In order to alleviate the model's dependency on labeled data, reduce annotation workload, and fully exploit road characteristics, we proposed a semi-supervised road extraction network via multiple consistency and multitask constraints (MCMCNet) that utilizes only minimal labeled data, while exploiting unlabeled data through the mining of pseudo-label information for constraint. Moreover, to ensure the generation of more accurate pseudo-labels, we incorporated a guided contrastive learning module (GCLM) into the model to increase interclass discriminability and enhance consistency constraints. In addition, to ensure the continuity of road extraction and integrity of the main roads, we added a road skeleton (road centerline) prediction head (RSPH) in addition to the original road segmentation prediction head. Finally, we introduced an adaptive road augment module (ARAM) to enhance linear road features and avoid learning redundant information by the use of local and global information adapted to road features. Extensive experiments demonstrated that MCMCNet achieved a 3%-5% improvement in $F1$ and IoU across three benchmark datasets, compared to other classical semi-supervised road extraction models, and the visualization results confirmed that MCMCNet partially addressed challenges including road occlusion, foreground-background high-similarity regions at extremely low label rates. The code is available at https://github.com/zhouyiqingzz/MCMCNet.

Improved MobileNet V3-Based Identification Method for Road Adhesion Coefficient

Road Adhesion Coefficient Estimation Based on Vehicle-Road Coordination and Deep Learning

Estimation of Tire-Road Peak Adhesion Coefficient for Intelligent Electric Vehicles Based on Camera and Tire Dynamics Information Fusion

Semantic Segmentation-Inspired Prediction of Pavement Adhesion Coefficient

Road adhesion coefficient Estimation: Physics-informed deep learning method with vehicle dynamics model

The Improved Algorithm for Identifying the Vehicle Road Adhesion Coefficient

Tire-Road Peak Adhesion Coefficient Estimation Method Based on Fusion of Vehicle Dynamics and Machine Vision.

Road Condition Estimation for Automotive Anti-Skid Control System Based on BP Neural Network

Improved AEB Algorithm Combined with Estimating the Adhesion Coefficient of Road Ahead and Considering the Performance of EHB

Road Friction Estimation Method Based on Fusion of Machine Vision and Vehicle Dynamics

Calculation Model for Adhesion Coefficient Based on Neural Networks System

MCMCNet: A Semi-supervised Road Extraction Network for High-resolution Remote Sensing Images Via Multiple Consistency and Multi-task Constraints

Integrating spatial and channel attention mechanisms with domain knowledge in convolutional neural networks for friction coefficient prediction

Adhesion Coefficient Identification of Wheeled Mobile Robot under Unstructured Pavement

Estimation of the Pavement Skid Resistance by Automated Vehicle Based on One-Dimensional Convolutional Neural Network

Reliable Identification of Road Surface Condition Considering Shadow Interference.

Estimation Method of Road Adhesion Coefficient for Vehicle Longitudinal Safety Assistant System

A comparative study of the recursive least squares and fuzzy logic estimation methods for the measurement of road adhesion coefficient

Robust Identification of Road Surface Condition Based on Ego-Vehicle Trajectory Reckoning

A Comprehensive Implementation of Road Surface Classification for Vehicle Driving Assistance: Dataset, Models, and Deployment

Multi-Sensor Fusion Based Estimation of Tire-Road Peak Adhesion Coefficient Considering Model Uncertainty