Search

The field of urban morphology, crucial for understanding the evolutionary trajectories of cityscapes, has traditionally depended on manual classification methods. The surge in deep learning and computer vision technologies presents an opportunity to automate and enhance urban typo-morphology studies. This research addresses three critical shortcomings in the current body of work: the neglect of urban fabric's three-dimensional qualities, the homogeneity of spatial scales in dataset creation and the dependence on a single-perspective for urban fabric classification. A novel deep learning-based model, UrbanClassifier, is introduced, trained on a substantial dataset that encapsulates the three-dimensionality of urban fabric along with morphological types and development periods. Extensive experimentation across four European cities highlights the model's ability to incorporate diverse spatial scales and viewpoints in urban fabric analysis. The UrbanClassifier exemplifies a method integrating features from various scales and perspectives, thus laying the groundwork for scalable and accessible urban typo-morphology studies, aiding practitioners in discerning the spatio-temporal evolution of urban fabric. • Merges open-source data for detailed 3D urban morphology and historical period labels. • Innovative assimilation of multi-scale, multi-view features for classifying urban typologies. • Facilitates scalable studies, aiding visualization of global urban growth patterns. [ABSTRACT FROM AUTHOR]

Deep learning applications in shaping ad hoc planning proposals are limited by the difficulty of integrating professional knowledge about cities with artificial intelligence. We propose a novel, complementary use of deep neural networks and planning guidance to automate street network generation that can be context-aware, learning-based, and user-guided. The model tests suggest that the incorporation of planning knowledge (e.g., road junctions and neighborhood types) in the model training leads to a more realistic prediction of street configurations. Furthermore, the new tool provides both professional and lay users an opportunity to systematically and intuitively explore benchmark proposals for comparisons and further evaluations. [ABSTRACT FROM AUTHOR]

Deep learning has attracted attention widely as the successful application of deep learning for vision tasks, such as image classification, object detection and so on. Due to the robustness and universality of deep learning, automotive manufacturing, a crucial part of national economy, needs deep learning to make production lines more intelligent and improve efficiency. However, some superior generally deep learning models, such as ViT, TNT, and Swin transformer, cannot meet automotive manufacturing requirements with high accuracy on a specific scene. As for automotive production lines, engineers usually adopt some smart designs, which can provide prior knowledge for designing deep learning models. Specifically, in an image, the position of target is usually fixed. Therefore, in order to take advantage of prior position, this paper designs a local mixer with prior position to capture local feature. Its main idea is that dividing the whole feature map into window feature maps and connecting window feature maps along channel dimension in order to make convolution kernel parameters for each window feature map are independent from others. Besides, MLP is adopted as global mixer to capture global feature and the pyramidal architecture with CNN is adopted. Comprehensive results demonstrate the effectiveness of proposed model on cars' type recognition. In particular, the proposed model achieves 97.938% accuracy on our data set, surpassing some transformer-like models. [ABSTRACT FROM AUTHOR]

Limited attention has been given to human-computer interactions in the plan-making process to capitalize on the relative strengths of both. This paper proposes a methodological framework for an interactive street network design that complements user-driven (i.e., procedural-based tools) and example-driven (i.e., learning-based tools) approaches in urban planning and design. The proposed framework consists of three components: (1) a data preparation module to link open-source road networks with human-labeled planning guidance, (2) a multi-stage deep learning (MSDL) model to reinforce the user-defined guidance in the automatic generation of street networks, and (3) a human-computer interaction (HCI) interface to enable the progressive design process. The performance and the working mechanism of the proposed framework were examined through experiments in four European cities (i.e., Amsterdam, Barcelona, Berlin, and Prague). The experiments demonstrate that the proposed MSDL model can achieve a better predictive performance compared to benchmark models, particularly when limited planning guidance is given. These finding are revealed using either computer vision- or street network-related metrics. With less than 40% of the ground truth planning guidance used as an input, the MSDL model can perform as well as other models using 100% of the information. Furthermore, when embedded within an HCI system for user trials, the model can facilitate a human-computer collaborative design process. This advantage is derived from the model's ability to provide initial prototypes, timely responses to changed guidance, and quantifiable evaluations of the generated proposals. Suitable for professionals and laypersons, the proposed tool can inform plan-making and public engagement by offering realistic, enriched, and human-centered spatial proposal alternatives for comparison. • Complement the procedural- and learning-based computer-aided frameworks for street network design. • Improve conventional, end-to-end deep-learning structure to a multi-stage one. • Develop an HCI system for interactive and progressive street network design. • Provide more realistic planning options and enrich alternatives in the early stages of urban planning and design. [ABSTRACT FROM AUTHOR]