Applications of Street View Images based on Artificial Intelligence: A Comprehensive Survey

AN ATTENTION-INTEGRATED YOLO11-S MODEL FOR OBJECT DETECTION IN SATELLITE IMAGERY: PERFORMANCE ANALYSIS AND COMPARISON

As remote sensing technologies improve, we are now able to look at the Earth from different points of view. These technologies have enabled major changes in many areas. High-resolution satellite images have enabled progress in civilian and military applications such as environmental monitoring, disaster management, and public safety. While these images give us much information, we also need advanced analysis and automated object detection. Recent studies employing deep learning architectures reached high success rates in object detection from satellite imagery. This study introduces a novel model by assessing the impacts of several YOLO object detection algorithms with the Convolutional Block Attention Module (CBAM) on aircraft detection from satellite images. We used the HRPlanesv2 dataset for our experiments. The results revealed that the proposed model had better performance compared to alternative models. The proposed model achieved a mAP50 of 0.9868, a mAP50-95 of 0.7911, a precision of 0.9811, and a recall of 0.9637. Incorporating CBAM improves the detection of objects in crowded and complex scenes. These results demonstrate that attention mechanisms have a significant impact when used with the YOLO architecture for object detection in satellite images. It also provides a reliable and efficient solution for practical applications requiring accurate and consistent aircraft detection.

Quantitative Perception of Street Space Based on Deep Learning and Future Morphology Prediction Using AIGC Technology

Street space facilitates public life within communities, and advancements in artificial intelligence technology bring innovative approaches to assessing and forecasting street space perceptions. This study is centered on the Tsim Sha Tsui area of Hong Kong, as an illustrative case. Street view images and historical photographs are used as primary data sources, leveraging deep learning for semantic segmentation to measure the constituent elements of street space. A perceptual evaluation model is developed using regression analysis and time-series forecasting models to quantify and describe in detail the regional styles and their evolution throughout the historical development of urban streets. The findings indicate that architectural elements consistently account for an average of 42.3% historically, while the proportion of carriageways has significantly increased since the 1980s, rising by 1.8% annually. Although the rate of greening has risen, sky visibility has declined, leading to a decreased level of openness. Utilizing ordinary least squares regression in conjunction with long short-term memory time-series predictions, it is anticipated that the share of buildings and signage will reach 57.6% by 2050 (+15.3% compared with 2020), with the share of sky and greenery decreasing to 18.9%. Subsequently, by integrating the quantitative results and substituting relevant assumptions, projections regarding the future transformations of street styles are made. Artificial intelligence–generated content (AIGC) technology is applied to adjust weights, enabling the fusion and transition of various elemental styles. The Stable Diffusion model is employed to generate three images representing future styles (traffic city, green city, and consumer city), and the correlation between the quantitative indicators and style migration is validated. This study demonstrates that the methodology employed can effectively analyze the historical evolution of street space and forecast multiple future scenarios, thereby providing data support for the revitalization of high-density urban streets. On the one hand, it can inform planning decisions by establishing thresholds for elemental occupancy, and on the other hand, it can assist in the selection of various scenarios through stylized image generation. Implementing street space quantification techniques and forecasting future styles provide valuable support for designers and evaluators of urban street design schemes, providing crucial insights for the design of street space and architectural form.

Scene Classification in Rural Environments for Autonomous Driving Using U-Net and ResNet

This paper addresses the semantic segmentation of road images, a critical task for autonomous vehicle navigation, particularly in non-urban environments that present significant challenges. While much research focuses on well-maintained roads in developed countries, this study confronts the complexities of real-world conditions, such as those prevalent in developing nations, which feature vast networks of unpaved and poorly maintained roads. The core of our methodology is a neural network architecture that synergistically combines the encoder-decoder structure of U-Net with the feature extraction power of a ResNet backbone. The primary objective is the precise classification of each image pixel into one of four essential categories for navigation: background, asphalt, paved, and unpaved road. The model's training regimen involved exploring different ResNet versions (ResNet18, ResNet34, and ResNet50) as the encoder backbone to assess the impact of network depth. A key aspect of our approach was a progressive training strategy, where model versions were trained on images of varying resolutions. The results demonstrated a significant and somewhat counter-intuitive finding: training the ResNet34-U-Net model with images at half the original resolution yielded the best overall performance, achieving the highest Dice and IoU scores. This suggests that reducing image resolution acts as an effective form of regularization, compelling the model to learn more general and robust features by ignoring minor, irrelevant details. This outcome not only enhances the model's generalization capabilities for diverse and imperfect road conditions but also carries a substantial practical advantage by reducing the computational cost of training and inference, a crucial factor for deployment on resource-constrained embedded systems in autonomous vehicles.

LLFA-Net: a collaborative CNN-transformer framework for land cover classification using high-resolution remote sensing images

A Hybrid Pyramid and Strip Pooling Network for Accurate Building Extraction from Remote Sensing Images

SRB-Net is presented, a U-Net-based framework that combines three complementary components: strip pooling for long-range horizontal and vertical context; residual multi-scale atrous spatial pyramid pooling with squeeze-and-excitation blocks for multi-scale and channel-aware feature learning; and a bottleneck attention module (BAM) for refining skip-connection features.

ConvNeXt-Tiny for High-Accuracy Natural Scene Image Classification with Test-Time Augmentation