Road pavement cracks automated detection is one of the key factors to evaluate the road distress quality, and it is a difficult issue for the construction of intelligent maintenance systems. However, pavement cracks automated detection has been a challenging task, including strong nonuniformity, complex topology, and strong noise-like problems in the crack images, and so on. To address these challenges, we propose the CrackSeg—an end-to-end trainable deep convolutional neural network for pavement crack detection, which is effective in achieving pixel-level, and automated detection via high-level features. In this work, we introduce a novel multiscale dilated convolutional module that can learn rich deep convolutional features, making the crack features acquired under a complex background more discriminant. Moreover, in the upsampling module process, the high spatial resolution features of the shallow network are fused to obtain more refined pixel-level pavement crack detection results. We train and evaluate the CrackSeg net on our CrackDataset, the experimental results prove that the CrackSeg achieves high performance with a precision of 98.00%, recall of 97.85%, F-score of 97.92%, and a mIoU of 73.53%. Compared with other state-of-the-art methods, the CrackSeg performs more efficiently, and robustly for automated pavement crack detection.

Road extraction is one of the most significant tasks for modern transportation systems. This task is normally difficult due to complex backgrounds such as rural roads that have heterogeneous appearances with large intraclass and low interclass variations and urban roads that are covered by vehicles, pedestrians and the shadows of surrounding trees or buildings. In this paper, we propose a novel method for extracting roads from optical satellite images using a refined deep residual convolutional neural network (RDRCNN) with a postprocessing stage. RDRCNN consists of a residual connected unit (RCU) and a dilated perception unit (DPU). The RDRCNN structure is symmetric to generate the outputs of the same size. A math morphology and a tensor voting algorithm are used to improve RDRCNN performance during postprocessing. Experiments are conducted on two datasets of high-resolution images to demonstrate the performance of the proposed network architectures, and the results of the proposed architectures are compared with those of other network architectures. The results demonstrate the effective performance of the proposed method for extracting roads from a complex scene.

Road detection is a focus of research in the field of remote sensing image analysis. This task is normally difficult due to the complexity of the data, which are heterogeneous in appearance with large intra-class and lower inter-class variations that frequently lead to large numbers of gaps and errors in road extraction. In this paper, a novel road detection method is proposed that combines a multiscale deep residual convolution neural network (MDRCNN) with postprocessing. The MDRCNN is used to obtain road areas more accurately and quickly. Multiscale convolution, which provides greater accuracy, is used to acquire the hierarchical features of different dimensions. The residual connections and global average pooling are introduced to improve the efficiency of the network in the process of backpropagation and forward propagation, respectively. In the postprocessing stage, the centerline of the road can be obtained based on the road area. Geometric constraints and mathematical morphological refinement as well as leaf-to-leaf connection are used to obtain the road line. Rectangular buffer analysis and a sector descriptor tracking connection are subsequently used to improve the integrity and accuracy of the road. We experimented on two datasets with different resolutions and different scenes. Compared with other neural network methods, our method is better at connecting road fractures and eliminating errors.

Cross-scene hyperspectral image classification has achieved favorable outcomes in the domain adaptation of deep learning. However, transferring the sample features learned from the source domain to the target domain remains challenging due to the interdomain differences. To address this issue, we propose a multi-domain adaptive unsupervised learning algorithm based on a generative network (MDAULA), which comprises three core components. First, the architecture consists of two branches, representing source domain(SD) and target domain(TD), respectively. Each branch is embedded with a generation module based on contrastive learning in order to form the generation domain(GD). GD is composed of the spectral and spatial features obtained by decoupling the original data (Source domain or target domain). GD is fused with the original data of the corresponding branch to form a mixed domain(MD). GD and MD are designed to minimize the differences of domain adaptation during the prediction process. Thus, GD, MD, the source domain (SD), and the target domain (TD) are simultaneously learned, enabling multi-domain integration. Second, we introduce a novel domain adaptive method for multi-level feature correlation alignment loss(CAL). It can orderly and hierarchically align the means and covariances of the features in the source and target domain, explicitly model the fine-grained distribution differences between feature channels, so that the model can balance the importance of features at different levels adaptively. Finally, we design an evaluation strategy for the most uncertain samples in the target domain. By calculating the entropy of these samples, we identify difficult samples and assign corresponding weights to further train and fine-tune the model. To validate the effectiveness of the proposed method, we conducted a series of comparative and ablation experiments. Experimental results on three cross-scene datasets demonstrate that MDAULA outperforms existing domain adaptation methods.

Road traffic network (RTN) structure plays an important role in the field of complex network analysis. In this paper, we propose a regional patch detection method from RTN via community detection of complex network. Firstly, the refined Adapted PageRank algorithm, which combines with the influence factors of the location property weight, the geographic distance weight and the road level weight, is used to calculate the candidate ranking results of key nodes in the RTN. Secondly, the ranking result and the shortest path distance as two significant impact factors are used to select the key points of the RTN, and then the Adapted K-Means algorithm is applied to regional patch detection of the RTN. Finally, based on the experimental data of Zhangwu road traffic network, the analysis results are as follows: Zhangwu is divided into 9 functional structures with key node locations as the core. Regional patch structure is divided according to key points, and the RTN is actually divided into nine small functional communities. Nine functional regional patches constitute a new network structure, maintaining connectivity between the regional patches can improve the overall efficiency of the RTN.