Road-surface identification is important for transportation monitoring and maintenance. However, this task is challenging due to the complexity of vibration signals, feature overlap among different surface types, and variations in real-world operating conditions. These challenges become more significant in time-series classification, where models must achieve high accuracy while remaining computationally efficient and suitable for low-cost hardware. This study investigates the design and evaluation of an automatic road-surface classification system using motion data collected from inertial sensors mounted on a vehicle, including accelerometers and gyroscopes. The system segments synchronized IMU signals into fixed-length windows and assigns each window to a predefined road-surface category. To address this problem, this study proposes CGB-Net, a lightweight and efficient deep learning architecture for road-surface classification. In this architecture, the integration of modules is designed to capture hierarchical features: one-dimensional convolutional neural networks (1D-CNN) are used to extract local temporal features from accelerometer and gyroscope signals, Gated Recurrent Units (GRU) model short-term temporal dependencies, and Bidirectional Long Short-Term Memory (Bi-LSTM) networks capture global long-term temporal context in both directions. The model is trained to distinguish among three types of road surfaces representing different material properties and degradation states: Asphalt_10 (new asphalt, less than 10 years old), Asphalt_15 (aged asphalt, more than 15 years old), and Concrete. Experimental results show that the proposed system achieves high classification performance, with both accuracy and F1-score exceeding 95%. These results indicate strong potential for practical applications in automated road monitoring and maintenance systems.

This study presents CGB-Net, a novel deep learning architecture specifically developed for classifying twelve distinct sleep positions using a single abdominal accelerometer, with direct applicability to gastroesophageal reflux disease (GERD) monitoring. Unlike conventional approaches limited to four basic postures, CGB-Net enables fine-grained classification of twelve clinically relevant sleep positions, providing enhanced resolution for personalized health assessment. The architecture introduces a unique integration of three complementary components: 1D Convolutional Neural Networks (1D-CNN) for efficient local spatial feature extraction, Gated Recurrent Units (GRU) to capture short-term temporal dependencies with reduced computational complexity, and Bidirectional Long Short-Term Memory (Bi-LSTM) networks for modeling long-term temporal context in both forward and backward directions. This complementary integration allows the model to better represent dynamic and contextual information inherent in the sensor data, surpassing the performance of simpler or previously published hybrid models. Experiments were conducted on a benchmark dataset consisting of 18 volunteers (age range: 19–24 years, mean 20.56 $\pm$ 1.1 years; height 164.78 $\pm$ 8.18 cm; weight 55.39 $\pm$ 8.30 kg; BMI 20.24 $\pm$ 2.04), monitored via a single abdominal accelerometer. A subject-independent evaluation protocol with multiple random splits was employed to ensure robustness and generalizability. The proposed model achieves an average Accuracy of 87.60% and F1-score of 83.38%, both reported with standard deviations over multiple runs, outperforming several baseline and state-of-the-art methods. By releasing the dataset publicly and detailing the model design, this work aims to facilitate reproducibility and advance research in sleep posture classification for clinical applications.