Industrial Internet of Things (IIoT) networks, as the application of IoT networks to modern industry, are growing rapidly with the digital transformation accelerates. Mobile sensors are widely adopted in monitoring key information of massive IIoT networks in an ad hoc fashion or retrofitting on to an existing infrastructure. Hence, owing to the characteristics of scalability, unstable locations and highly unstructured, achieving perfect prediction for mobile multi-sensor is a challenging problem. In this paper, a dynamic space-time prediction algorithm combining temporal convolutional network and graph convolutional network (TCN-GCN) is proposed to provide reliable multiple nodes prediction for detection and maintenance in IIoT networks. In particularly, an adaptive learning graph process is exclusively designed according to the dynamic characteristics of mobile sensors. Moreover, the improved dilated time convolution network and dynamic graph convolution network are combined to effectively capture the time dependence and topology information. Furthermore, an advanced loss function is applied to avoid overfitting. Numerical simulations show that the proposed TCN-GCN prediction algorithm exhibits higher effectiveness in accuracy and complexity than convolutional long short-term memory (ConvLSTM), temporal graph convolutional network (T-GCN) and residual graph convolutional long short-term memory (Graph-ResLSTM).

Researchers in the fields of ecological environment and remote sensing pay considerable attention to the estimation of forest canopy height with synthetic aperture radar (SAR). The interest is due to the ability of SAR to penetrate the forest canopy and its sensitivity to forest properties through backscattered intensity. Recent advances in deep learning (DL) present the possibility to derive canopy height maps from single high-resolution (HR) SAR images using neural networks. SAR2Canopy, an innovative framework, is proposed in this paper for canopy height estimation based, which incorporates sensor and scattering knowledge into the estimation. The integration framework allows for the merging of reconstruction trunk scattering features by an equivalent dihedral corner reflector (DCR) scattering model into the supervised tree height estimation process. The proposed method attempts a new approach combining the physical characteristics of SAR data with the nonlinear feature learning ability of DL, potentially extending to different DL algorithms. Experiments are conducted with airborne fully polarimetric P-band SAR data from two areas. Compared to the baseline models, including UNet, DeepLabV3_ResNet50, and FCN_ResNet50, the proposed SAR2Canopy integrated with the DCR model achieves an increase of up to 13.66% in the coefficient of determination (R²), while reducing the root mean squared error (RMSE) and mean absolute error (MAE) by as much as 14.59% and 20.69%, respectively.

Forest above-ground biomass (AGB) is essential for understanding forest carbon stocks and ecosystem dynamics. Despite the strong potential of synthetic aperture radar (SAR) for forest monitoring, retrieving AGB from SAR observations remains difficult because the relationship between SAR scattering responses and forest biomass is highly nonlinear, spatially variable, and often difficult to represent with both accuracy and interpretability. To address this issue, this study proposes PhySR-Bio, a multi-stage physics-aware symbolic regression framework for interpretable SAR-based forest biomass retrieval. The physics-aware design is achieved by incorporating SAR scattering-related variables, incidence angle information, and physically meaningful feature combinations into symbolic regression. Unlike black box regression models, PhySR-Bio derives an analytical formula in which the selected variables and their mathematical relationships can be directly inspected. On the independent test set, PhySR-Bio achieved an RMSE of 45.083 Mg/ha, an MAE of 33.997 Mg/ha, and an R of 0.527. It reduced RMSE and MAE by up to 6.35% and 8.57%, respectively, compared with representative semi-empirical formulas, and achieved a test RMSE within 2.3% of the best-performing black box machine learning model while retaining an explicit and interpretable expression.