Search for: All records

Extracting roads in aerial images has numerous applications in artificial intelligence and multimedia computing, including traffic pattern analysis and parking space planning. Learning deep neural networks, though very successful, demands vast amounts of high-quality annotations, of which acquisition is time-consuming and expensive. In this work, we propose a semi-supervised approach for image-based road extraction where only a small set of labeled images are available for training to address this challenge. We design a pixel-wise contrastive loss to self-supervise the network training to utilize the large corpus of unlabeled images. The key idea is to identify pairs of overlapping image regions (positive) or non-overlapping image regions (negative) and encourage the network to make similar outputs for positive pairs or dissimilar outputs for negative pairs. We also develop a negative sampling strategy to filter false negative samples during the process. An iterative procedure is introduced to apply the network over raw images to generate pseudo-labels, filter and select high-quality labels with the proposed contrastive loss, and re-train the network with the enlarged training dataset. We repeat these iterative steps until convergence. We validate the effectiveness of the proposed methods by performing extensive experiments on the public SpaceNet3 and DeepGlobe Road datasets. Results show that our proposed method achieves state-of-the-art results on public image segmentation benchmarks and significantly outperforms other semi-supervised methods.

 

Summary China’s Belt and Road Initiative (BRI), designed to build critical infrastructure and coordinate economic growth, is the most significant development initiative in modern history. The BRI has a documented vision for sustainability, including environmental impact assessments and responsibility tenets. Despite this, a growing body of literature has found adverse effects of BRI projects on protected land and species. To understand corporate responsibility and regulations for companies participating in the BRI, we gathered information on 260 BRI companies using the Refinitiv Eikon BRI Connect database and the China Global Investment Tracker. The results revealed a significant gap in corporate responsibility reporting for biodiversity impacts, environmental restoration, environmental project financing and the United Nations’ Sustainable Development Goals (SDG) 14 ‘Life below Water’ and 15 ‘Life on Land’. The modest fraction of companies that we found to report biodiversity accountability highlights the need to restructure and incentivize the reporting of environmental and biodiversity risks. The current evidence of limited adherence to responsibility measures highlights a clear opportunity to align BRI development with the BRI’s vision for sustainability, and to strengthen links for policy engagement within Chinese regulatory frameworks and international obligations at the United Nations within its SDG framework. 

Detecting small objects (e.g., manhole covers, license plates, and roadside milestones) in urban images is a long-standing challenge mainly due to the scale of small object and background clutter. Although convolution neural network (CNN)-based methods have made significant progress and achieved impressive results in generic object detection, the problem of small object detection remains unsolved. To address this challenge, in this study we developed an end-to-end network architecture that has three significant characteristics compared to previous works. First, we designed a backbone network module, namely Reduced Downsampling Network (RD-Net), to extract informative feature representations with high spatial resolutions and preserve local information for small objects. Second, we introduced an Adjustable Sample Selection (ADSS) module which frees the Intersection-over-Union (IoU) threshold hyperparameters and defines positive and negative training samples based on statistical characteristics between generated anchors and ground reference bounding boxes. Third, we incorporated the generalized Intersection-over-Union (GIoU) loss for bounding box regression, which efficiently bridges the gap between distance-based optimization loss and area-based evaluation metrics. We demonstrated the effectiveness of our method by performing extensive experiments on the public Urban Element Detection (UED) dataset acquired by Mobile Mapping Systems (MMS). The Average Precision (AP) of the proposed method was 81.71%, representing an improvement of 1.2% compared with the popular detection framework Faster R-CNN. 

Although research on wildlife species across taxa has shown that males and females may differentially select habitat, sex-specific habitat suitability models for endangered species are uncommon. We developed sex-specific models for Bengal tigers (Panthera tigris) based on camera trapping data collected from 20 January to 22 March 2010 within Chitwan National Park, Nepal, and its buffer zone. We compared these to a sex-indiscriminate habitat suitability model to assess the benefits of a sex-specific approach to habitat suitability modeling. Our sex-specific models produced more informative and detailed habitat suitability maps and highlighted vital differences in the spatial distribution of suitable habitats for males and females, specific associations with different vegetation types, and habitat use near human settlements. Improving and refining habitat models for this and other critically endangered species provides the necessary information to meet established conservation goals and population recovery targets. 

Community forests have been established worldwide to sustainably manage forest ecosystem services while maintaining the livelihoods of local residents. The Chitwan National Park in Nepal is a world-renowned biodiversity hotspot, where community forests were consolidated in the park’s buffer zone after 1993. These western Chitwan community forests stand as the frontiers of human–environment interactions, nurturing endangered large mammal species while providing significant natural resources for local residents. Nevertheless, no systematic forest cover assessment has been conducted for these forests since their establishment. In this study, we examined the green vegetation dynamics of these community forests for the years 1988–2018 using Landsat surface reflectance products. Combining an automatic water extraction index, spectral mixture analysis and the normalized difference fraction index (NDFI), we developed water masks and quantified the water-adjusted green vegetation fractions and NDFI values in the forests. Results showed that all forests have been continuously greening up since their establishment, and the average green vegetation cover of all forests increased from approximately 30% in 1988 to above 70% in 2018. With possible contributions from the invasion of exotic understory plant species, we credit community forestry programs for some of the green-up signals. Monitoring of forest vegetation dynamics is critical for evaluating the effectiveness of community forestry as well as developing sustainable forest management policies. Our research will provide positive feedbacks to local community forest committees and users.