Road extraction from high-resolution remote sensing images (HRSI) is a challenging but hot research topic in the past decades. A large number of methods are invented to deal with this problem. This article provides a comprehensive review of these existing approaches. We classified the methods into heuristic and data-driven. The heuristic methods are the mainstream in the early years, and the data-driven methods based on deep learning have been quickly developed recently. With regard to the heuristic methods, the road feature model is firstly introduced, then, the classic extraction methods are reviewed in two sub-categories: semi-automatic and automatic. The principles, inspirations, advantages and disadvantages of these methods are described. In terms of the data-driven methods, the road extraction methods based on deep neural network, particularly those based on patched convolutional neural network, fully convolutional network and generative adversarial network are reviewed. We perform subjective comparisons between the methods inner each type. Furthermore, the quantity performances achieved on the same dataset are compared between the heuristic and data-driven methods to show the strengthening of the data-driven methods. Finally, the conclusion and prospects are summarized.

Integrating Tiny Machine Learning (TinyML) with edge computing in remotely sensed images enhances the capabilities of road anomaly detection on a broader level. Constrained devices efficiently implement a Binary Neural Network (BNN) for road feature extraction, utilizing quantization and compression through a pruning strategy. The modifications resulted in a 28-fold decrease in memory usage and a 25% enhancement in inference speed while only experiencing a 2.5% decrease in accuracy. It showcases its superiority over conventional detection algorithms in different road image scenarios. Although constrained by computer resources and training datasets, our results indicate opportunities for future research, demonstrating that quantization and focused optimization can significantly improve machine learning models’ accuracy and operational efficiency. ARM Cortex-M0 gives practical feasibility and substantial benefits while deploying our optimized BNN model on this low-power device: Advanced machine learning in edge computing. The analysis work delves into the educational significance of TinyML and its essential function in analyzing road networks using remote sensing, suggesting ways to improve smart city frameworks in road network assessment, traffic management, and autonomous vehicle navigation systems by emphasizing the importance of new technologies for maintaining and safeguarding road networks.

To detect lanes at night, each detecting image is the fusion of the multiple images in a video sequence. The valid lane line detection region is identified on region merging. Then, the image preprocessing algorithm based on the Fragi algorithm and Hessian matrix is applied to enhance lanes; to extract the lane line center feature points, the image segmentation algorithm based on Fractional differential is proposed; and according to the possible lane line positions, the algorithm detects the centerline points in four directions. Subsequently, the candidate points are determined, and the recursive Hough transformation is applied to obtain the possible lane lines. Finally, to obtain the final lane lines, we assume that one lane line should have an angle between 25 and 65 degrees, while the other should have an angle between 115 and 155 degrees, if the detected line is not in the regions, the Hough line detection will be continued by increasing the threshold value until the two lane lines are got. By testing more than 500 images and comparing deep learning methods and image segmentation algorithms, the lane detection accuracy by the new algorithm is up to 70%.