With the tremendous growth in online information, capturing dynamic users’ preferences based on their historical interactions and providing a few desirable items to users have become an urgent service for all businesses. Recurrent neural networks (RNNs) and item-based collaborative filtering (CF) models are commonly used in industries due to their simplicity and efficiency. However, they fail to different contexts that could differently influence current users’ decision-making. Also, they are not sufficient to capture multiple users’ interests based on features of the interacting items. Besides, they have a limited modeling capability for the evolution of diversity and dynamic user preferences. In this article, we exploit long- and short-term preferences for deep context-aware recommendations (LSCAR) to enhance the next item recommendation’s performance by introducing three novel components as follows: 1) the user–contextual interaction module is proposed to capture and differentiate the interaction between contexts and users; 2) the encoded multi-interest module is introduced to capture various types of user interests; and 3) the integrator fusion gate module is used to effectively fuse the related long-term interests to the current short-term part, and the module returns the final user interest representation. Extensive experiments and results for two public datasets demonstrate that the proposed LSCAR outperforms the state-of-the-art models in almost all metrics and could provide interpretable recommendation results.
Vehicle speed estimation is one of the most critical issues in intelligent transportation system (ITS) research, while defining distance and identifying direction have become an inseparable part of vehicle speed estimation. Despite the success of traditional and deep learning approaches in estimating vehicle speed, the high cost of deploying hardware devices to get all related sensor data, such as infrared/ultrasonic devices, Global Positioning Systems (GPS), Light Detection and Ranging (LiDAR systems), and magnetic devices, has become the key barrier to improvement in previous studies. In this paper, our proposed model consists of two main components: 1) a vehicle detection and tracking component – this module is designed for creating reliable detection and tracking every specific object without doing calibration; 2) homography transformation regression network – this module has a function to solve occlusion issues and estimate vehicle speed accurately and efficiently. Experimental results on two datasets show that the proposed method outperforms the state-of-the-art methods by reducing the mean square error (MSE) metric from 14.02 to 6.56 based on deep learning approaches. We have announced our test code and model on GitHub with https://github.com/ervinyo/Speed-Estimation-Using-Homography-Transformation-and-Regression-Network.
Human action recognition (HAR) is growing in machine learning with a wide range of applications. One challenging aspect of HAR is recognizing human actions while playing music, further complicated by the need to recognize the musical notes being played. This paper proposes a deep learning-based method for simultaneous HAR and musical note recognition in music performances. We conducted experiments on Morin khuur performances, a traditional Mongolian instrument. The proposed method consists of two stages. First, we created a new dataset of Morin khuur performances. We used motion capture systems and depth sensors to collect data that includes hand keypoints, instrument segmentation information, and detailed movement information. We then analyzed RGB images, depth images, and motion data to determine which type of data provides the most valuable features for recognizing actions and notes in music performances. The second stage utilizes a Spatial Temporal Attention Graph Convolutional Network (STA-GCN) to recognize musical notes as continuous gestures. The STA-GCN model is designed to learn the relationships between hand keypoints and instrument segmentation information, which are crucial for accurate recognition. Evaluation on our dataset demonstrates that our model outperforms the traditional ST-GCN model, achieving an accuracy of 81.4%.
Human action recognition has recently attracted much attention in computer vision research. Its applications are widely found in video surveillance, human-computer interaction, entertainment, and autonomous driving. In this study, we developed a system for evaluating online music performances. This system conducts experiments to assess performance of playing the erhu, the most popular traditional stringed instrument in East Asia. Mastering the erhu poses a challenge, as players often struggle to enhance their skills due to incorrect techniques and a lack of guidance, resulting in limited progress. To address this issue, we propose hybrid models based on graph convolutional networks (GCN) and temporal convolutional networks (TCN) for action recognition to capture spatial relationships between different joints or keypoints in a human skeleton, and interactions between these joints. This can assist players in identifying errors while playing the instrument. In our research, we use RGB video as input, segmenting it into individual frames. For each frame, we extract keypoints, encompassing both image and keypoint information, which serve as input data for our model. Leveraging our innovative model architecture, we achieve an impressive accuracy rate exceeding 97% across various classes of hand error modules, thus providing valuable insights into the assessment of musical performances and demonstrates the potential of AI-based solutions to enhance the learning and correction of complex human actions in interactive learning environments.
Deep neural network-based tracking tasks have experienced significant advancements in recent years. However, these networks continue to face challenges in effectively adapting to appearance changes in both target and background, as well as linking objects after extended periods. The primary challenge in tracking lies in the frequent changes in a target’s appearance throughout the tracking process, which can potentially reduce tracker robustness when faced with issues such as aspect ratio changes, occlusions, scale variations, and confusion from similar objects. To address this challenge, we propose a tracking architecture that combines a temporal convolutional network (TCN) and attention mechanism with spatial-temporal memory. The TCN component empowers the model to capture temporal dependencies, while the attention mechanism reduces computational complexity by focusing on crucial regions based on context. We leverage the target’s historical information stored in the spatial-temporal memory network to guide the tracker in better adapting to target deformation. Our model attains a 63.0% average overlap (AO) on the GOT-10K dataset, a 71.9% success score (AUC) on OTB2015, a 65.6% success score (AUC) on UAV123, and achieves 56.9% accuracy on the VOT2018 dataset. These outcomes demonstrate the high effectiveness of our proposed tracker in tracking a single object.
29 September 2025 in Dalat University Journal of Science
In the past decade, significant progress has been made in hyperspectral image (HSI) classification due to advances in deep learning techniques. This study introduces a novel computational framework for classifying HSIs using a multiscale 3D convolutional neural network (3DCNN). The purpose of this research is to enhance the efficiency of HSI classification by leveraging the spatial-spectral characteristics of HSI data. The methodology involves preprocessing HSI datasets, extracting features using parallel 3DCNN layers with skip connections, and evaluating performance on five benchmark datasets. Key findings demonstrate that the proposed model significantly improves classification accuracy, achieving average accuracies of 85%, 84%, 94%, 98%, and 95% on the Kennedy Space Center (KSC), Indian Pines (IP), Pavia University (PU), Pavia Centre (PC), and Botswana (BW) datasets, respectively. The originality of this study lies in the multiscale parallel design of the 3DCNN architecture, which balances computational complexity with enhanced performance. Limitations include the high computational cost associated with training deep networks, which can be mitigated by graphics processing unit (GPU) acceleration and optimized implementations.