Temporal Entropy Hybrid Labelling (TEHL) With GCNN: An Intelligent Deep Learning Framework For Thz Satellite-Based Environmental Monitoring Using Carbon Nanocomposite Sensors

Abstract

Environmental monitoring requires intelligent sensing systems capable of analysing spatially distributed, time-varying, and uncertain environmental conditions. Conventional monitoring approaches often provide limited adaptive labelling and fail to capture spatial dependencies among distributed monitoring locations. To address these limitations, it proposes a Temporal Entropy Hybrid Labelling with Graph Convolutional Neural Network (TEHL-GCNN), a hybrid deep learning model for THz satellite-based environmental monitoring using carbon nanocomposite sensors. A dataset containing 4,320 environmental observations is constructed by integrating carbon nanocomposite sensor measurements with THz-derived and satellite-supported spatial, temporal, spectral, and environmental information. Savitzky–Golay filtering is applied to reduce signal noise while preserving signal characteristics, followed by Z-score normalization to standardize the observations. Node2Vec is then employed to extract low-dimensional node representations that capture structural relationships among monitoring locations. The TEHL mechanism generates adaptive environmental-condition labels by integrating temporal variation, information entropy, and environmental threshold characteristics, enabling improved representation of stable, transitional, and abnormal conditions. Subsequently, monitoring locations are represented as graph nodes, while spatial proximity, environmental similarity, and signal relationships define inter-node connections. The GCNN aggregates neighbourhood information to learn spatial dependencies and classify environmental conditions. The model is implemented using Python 3.11 and evaluated using classification metrics against conventional approaches. The proposed TEHL-GCNN achieves an accuracy of 96.86%, demonstrating its effectiveness in combining adaptive labelling with spatial dependency learning. Overall, the model provides a scalable approach for intelligent environmental monitoring using integrated THz, satellite, and carbon nanocomposite sensor observations.