The classification of urban land-use information has become the underlying database for a variety of applications including urban planning and administration. The lack of datasets and changeable semantics of land-use make deep learning methods suffer from low precision, which prevent improvements in the effectiveness of using AI methods for applications. In this paper, we first used GIS data to produce a well-tagged and high-resolution urban land-use image dataset. Then, we proposed a combined convolutional neural network named DUA-Net for complex and diverse urban land-use classification. The DUA-Net combined U-Net and Densely connected Atrous Spatial Pyramid Pooling (DenseASPP) to extract Remote Sensing Imagers (RSIs) features in parallel. Then, channel attention was used to efficiently fuse the multi-source semantic information from the output of the double-layer network to learn the association between different land-use types. Finally, land-use classification of high-resolution urban RSIs was achieved. Experiments were performed on the dataset of this paper, the publicly available Vaihingen dataset and Potsdam dataset with overall accuracy levels reaching 75.90%, 89.71% and 89.91%, respectively. The results indicated that the complex land-use types with heterogeneous features were more difficult to extract than the single-feature land-cover types. The proposed DUA-Net method proved suitable for high-precision urban land-use classification, which will be of great value for urban planning and national land resource surveying.

Existing remote sensing scene image (RSSI) classification methods usually rely on static closed-set assumption that testing samples do not belong to unknown classes. However, practical applications are usually the open-set classification problem, which means that RSSIs from unknown classes will appear in the testing set. Most existing methods are prone to forcibly misclassify RSSIs of unknown classes into known classes, resulting in poor practical performance. In this letter, a deep feature reconstruction learning (DFRL) framework is proposed for open-set classification of RSSIs. The proposed DFRL unifies discriminative feature learning and feature reconstruction into an end-to-end network. Firstly, a feature extraction module is utilized to project raw input data from the image space to the feature space to extract deep features. Then, the deep features are fed to a deep feature reconstruction module for distinguishing known and unknown classes based on feature-level reconstruction errors. The feature-level reconstruction can effectively suppress the interference of complex backgrounds. In addition, a sparse regularization is introduced to improve the discrimination of image representation. Experiments on three RSSI datasets demonstrate the effectiveness of DFRL for open-set classification of RSSIs.

Online learning and teaching have become the primary forms of education during the global pandemic, and online learning systems, which can provide fair educational opportunities for everyone, are increasingly important for sustainability in education. The amount of time a student spends on online learning systems affects the fairness and persistence of sustainability in education. To support personalized learning opportunities, interactive learning, and easy-to-access resources, we propose a map organization and visualization method called MapOnLearn for online learning systems. First, we converted tree-like hierarchical course units (HCUs) and knowledge points (KPs) into a fundamental two-dimensional (2D) map of hierarchically divided polygons and used the map to form containers to manage all learning resources. Then, we used the zoom feature of the map to express the hierarchical structures of knowledge and formulated corresponding rules for displaying information at different levels. Path analysis was applied to express the learning process, and topological processing was adapted to represent the relationships among HCUs and KPs. We developed maps for a high-level math course, a course on data structures, and an English course at a university in China and investigated 264 students and 27 teachers for a semester by using the technology acceptance model (TAM). We found that the map visualization and organization method had a positive impact on the way teachers and students use online learning systems and improved the online learning experience. To attract more students to spend more time on online learning, we hope our method can promote the sustainable development of education.

In recent years, with the rapid development of unmanned aerial vehicle (UAV) technology, low-altitude remote sensing imagery has been increasingly applied in urban planning, environmental monitoring, and disaster management. However, due to large variations in object scales and complex boundaries of ground objects, existing methods still face significant challenges in handling small targets and objects with intricate edges. To address these issues, this paper proposes a lightweight and efficient semantic segmentation network for UAV remote sensing images, named UAV-FAENet, aiming to improve the detection accuracy of small targets and the quality of edge segmentation. UAV-FAENet adopts an encoder-decoder architecture, integrating a High-Low frequency Adaptive Enhancement Module (HLAEM) and a decoder based on a state space modeling mechanism (CSRA-VSS decoder). Specifically, HLAEM decomposes features into low-frequency and high-frequency components in the frequency domain, enhances them separately, and dynamically fuses them through a channel attention mechanism to effectively preserve edge details. The CSRA-VSS decoder introduces a state space model to capture long-range dependencies in feature sequences and employs a cross-scale attention mechanism to enhance feature responses in small target regions. The proposed method achieves mIoU scores of 68.84%, 80.11%, and 70.26% on the UAVid, UDD6, and DroneDeploy UAV datasets, respectively. Our method outperforms existing mainstream approaches and demonstrates strong potential in improving small object recognition accuracy and edge detail recovery, providing a feasible solution for high-precision semantic segmentation in complex urban scenes.

Mining and predicting college students behaviors from fine-grained spatial-temporal campus activity data play key roles in the academic success and personal development of college students. Most of the existing behavior prediction methods use shallow learning algorithms such as statistics, clustering, and correlation analysis approaches, which fail to mine the long-term spatial-temporal dependencies and semantic correlations from these fine-grained campus data. We propose a novel multi-fragment dynamic semantic spatial-temporal graph convolution network, named the MFDS-STGCN, on the basis of a spatial-temporal graph convolutional network (STGCN) for the automatic prediction of college students' behaviors and abnormal behaviors. We construct a dataset including 7.6 million behavioral records derived from approximately 400 students over 140 days to evaluate the effectiveness of the prediction model. Extensive experimental results demonstrate that the proposed method outperforms multiple baseline prediction methods in terms of student behavior prediction and abnormal behavior prediction, with accuracies of 92.60% and 90.84%, respectively. To further enable behavior prediction, we establish an early warning management mechanism. Based on the predictions and analyses of Big Data, education administrators can detect undesirable abnormal behaviors in time and thus implement effective interventions to better guide students' campus lives, ultimately helping them to more effectively develop and grow.

The existing remote sensing image datasets target the identification of objects, features, or man-made targets but lack the ability to provide the date and spatial information for the same feature in the time-series images. The spatial and temporal information is important for machine learning methods so that networks can be trained to support precision classification, particularly for agricultural applications of specific crops with distinct phenological growth stages. In this paper, we built a high-resolution unmanned aerial vehicle (UAV) image dataset for middle-season rice. We scheduled the UAV data acquisition in five villages of Hubei Province for three years, including 11 or 13 growing stages in each year that were accompanied by the annual agricultural surveying business. We investigated the accuracy of the vector maps for each field block and the precise information regarding the crops in the field by surveying each village and periodically arranging the UAV flight tasks on a weekly basis during the phenological stages. Subsequently, we developed a method to generate the samples automatically. Finally, we built a high-resolution UAV image dataset, including over 500,000 samples with the location and phenological growth stage information, and employed the imagery dataset in several machine learning algorithms for classification. We performed two exams to test our dataset. First, we used four classical deep learning networks for the fine classification of spatial and temporal information. Second, we used typical models to test the land cover on our dataset and compared this with the UCMerced Land Use Dataset and RSSCN7 Dataset. The results showed that the proposed image dataset supported typical deep learning networks in the classification task to identify the location and time of middle-season rice and achieved high accuracy with the public image dataset.