Search

Road extraction from high-resolution remote sensing images is a critical task in image understanding and analysis, yet it poses significant challenges because of road occlusions caused by vegetation, buildings, and shadows. Deep convolutional neural networks have emerged as the leading approach for road extraction because of their exceptional feature representation capabilities. However, existing methods often yield incomplete and disjointed road extraction results. To address this issue, we propose CR-HR-RoadNet, a novel high-resolution road extraction network that incorporates local and global context reasoning. In this work, we introduce a road-adapted high-resolution network as the feature encoder, effectively preserving intricate details of narrow roads and spatial information. To capture multi-scale local context information and model the interplay between roads and background environments, we integrate multi-scale features with residual learning in a specialized multi-scale feature representation module. Moreover, to enable efficient long-range dependencies between different dimensions and reason the correlation between various road segments, we employ a lightweight coordinate attention module as a global context-aware algorithm. Extensive quantitative and qualitative experiments on three datasets demonstrate that CR-HR-RoadNet achieves superior extraction accuracy across various road datasets, delivering road extraction results with enhanced completeness and continuity. The proposed method holds promise for advancing road extraction in challenging remote sensing scenarios and contributes to the broader field of deep-learning-based image analysis for geospatial applications. [ABSTRACT FROM AUTHOR]

Building polygons plays an important role in urban management. Although leveraging deep learning techniques for building polygon extraction offers advantages, the models heavily rely on a large number of training samples to achieve good generalization performance. In scenarios with small training samples, the models struggle to effectively represent diverse building structures and handle the complexity introduced by the background. A common approach to enhance feature representation is fine-tuning a pre-trained model on a large dataset specific to the task. However, the fine-tuning process tends to overfit the model to the task area samples, leading to the loss of generalization knowledge from the large dataset. To address this challenge and enable the model to inherit the generalization knowledge from the large dataset while learning the characteristics of the task area samples, this paper proposes a knowledge distillation-based framework called Building Polygon Distillation Network (BPDNet). The teacher network of BPDNet is trained on a large building polygon dataset containing diverse building samples. The student network was trained on a small number of available samples from the target area to learn the characteristics of the task area samples. The teacher network provides guidance during the training of the student network, enabling it to learn under the supervision of generalization knowledge. Moreover, to improve the extraction of buildings against the backdrop of a complex urban context, characterized by fuzziness, irregularity, and connectivity issues, BPDNet employs the Dice Loss, which focuses attention on building boundaries. The experimental results demonstrated that BPDNet effectively addresses the problem of limited generalization by integrating the generalization knowledge from the large dataset with the characteristics of the task area samples. It accurately identifies building polygons with diverse structures and alleviates boundary fuzziness and connectivity issues. [ABSTRACT FROM AUTHOR]

Building extraction using deep learning techniques has advantages but relies on a large number of clean labeled samples to train the model. Complex appearance and tilt shots often cause many offsets between building labels and true locations, and these noises have a considerable impact on building extraction. This paper proposes a new knowledge distillation-based building extraction method to reduce the impact of noise on the model and maintain the generalization of the model. The method can maximize the generalizable knowledge of large-scale noisy samples and the accurate supervision of small-scale clean samples. The proposed method comprises two similar teacher and student networks, where the teacher network is trained by large-scale noisy samples and the student network is trained by small-scale clean samples and guided by the knowledge of the teacher network. Experimental results show that the student network can not only alleviate the influence of noise labels but also obtain the capability of building extraction without incorrect labels in the teacher network and improve the performance of building extraction. [ABSTRACT FROM AUTHOR]

Service recommendation is key to improving users' online experience. The development of the Internet has accelerated the creation of many services, and whether users can obtain good experiences among the massive number of services mainly depends on the quality of service recommendation. It is commonly believed that deep learning has excellent nonlinear fitting ability in capturing the complex interactions between users and items. The advantage in learning intricacy relationships enables deep learning to become an important technology for present service recommendation. Recently, it is noticed that linear models can perform almost as well as the state-of-the-art deep learning models, suggesting that capturing linear relationships between users and items is also very important for recommender systems. Therefore, numerous deep learning systems combined with linear models have been proposed. However, existing models are incapable of considering the size of the embedding. When the embedding dimension is too large, it leads to overfitting and thus influences the model's ability to capture linear relationships. In this paper, a neural network based on two-layer matrix factorization and multi-layer perceptron—Two-layer Matrix factorization and Multi-layer perceptron Neural Network (TMMNN)—is proposed. To solve the problem of overfitting caused by an oversized embedding dimension, multi-size embedding technology has been integrated into the model. Matrix factorization and the multi-layer perceptron are placed in the upper and lower layers respectively, and they both receive embedding vectors dynamically adjusted for dimensions. In the upper layer, the matrix factorization is responsible for receiving the embedding of users and items, capturing linear relationships, and then yielding the generated new vectors as input to the multi-layer perceptron in the lower layer. Compared to other previously proposed models, the experimental results on the standard datasets MovieLens 20M and MovieLens Latest show that the TMMNN model is evidently better in terms of prediction accuracy. [ABSTRACT FROM AUTHOR]

China's urbanization has dramatically accelerated in recent decades. Land for urban build-up has changed not only in large cities but also in small counties. Land cover mapping is one of the fundamental tasks in the field of remote sensing and has received great attention. However, most current mapping requires a significant manual effort for labeling or classification. It is of great practical value to use the existing low-resolution label data for the classification of higher resolution images. In this regard, this work proposes a method based on noise-label learning for fine-grained mapping of urban build-up land in a county in central China. Specifically, this work produces a build-up land map with a resolution of 10 m based on a land cover map with a resolution of 30 m. Experimental results show that the accuracy of the results is improved by 5.5% compared with that of the baseline method. This notion indicates that the time required to produce a fine land cover map can be significantly reduced using existing coarse-grained data. [ABSTRACT FROM AUTHOR]

The lack of pixel-level labeling limits the practicality of deep learning-based building semantic segmentation. Weakly supervised semantic segmentation based on image-level labeling results in incomplete object regions and missing boundary information. This paper proposes a weakly supervised semantic segmentation method for building detection. The proposed method takes the image-level label as supervision information in a classification network that combines superpixel pooling and multi-scale feature fusion structures. The main advantage of the proposed strategy is its ability to improve the intactness and boundary accuracy of a detected building. Our method achieves impressive results on two 2D semantic labeling datasets, which outperform some competing weakly supervised methods and are close to the result of the fully supervised method. [ABSTRACT FROM AUTHOR]

Due to the problem of complex aggregate stacking and adhesion, current construction site aggregate grade detection relies on traditional screening methods and single digital image processing technology, which causes inefficiency and segmentation identification difficulties. This problem has become a technical bottleneck in achieving automatic construction site mixed aggregate grade detection. This study constructs a noncontact testing platform for aggregate gradation based on a self-developed sampling and testing device for mixed aggregate gradation and improved deep learning algorithms, which enables rapid testing of mixed aggregate gradation. Based on the principle of instance segmentation, each aggregate in the stacked mixed aggregates collected based on hardware observation is used as an instance to detect independent aggregate targets in the mixed aggregate images with different moisture contents, sand contents and cohesive stacking. An improved aggregate segmentation convolutional neural network model (AS Mask RCNN: Aggregate Segmentation Mask RCNN) is used to achieve the gradation detection of mixed aggregates. This study employed three different types of experiments, and the results showed that the AS Mask RCNN network model achieved an accuracy of over 89.13% in the three experimental situations, and compared the results with those of the Faster RCNN and Mask R-CNN models, with an accuracy improvement of 8.85% and a reduction of 1.29 s in the processing time of a single image segmentation, which can meet the field near real-time detection requirements. The self-developed noncontact testing platform for aggregate grading can adapt to practical applications in complex environments, enabling digital, automated and intelligent noncontact rapid testing of mixed aggregate grading, further improving the accuracy of aggregate grading testing and serving the high-quality development of reservoir dam construction in China. • Aggregate stacking and adhesion affect the intelligent detection of particle size. • A noncontact inspection platform for aggregate grading based on deep learning. • Three different types of comparative experiments were used in this study. • An improved aggregate segmentation convolutional neural network model is proposed. • The study can be applied to actual engineering rockfill material gradation testing. [ABSTRACT FROM AUTHOR]