Hua Li

Human action recognition (HAR) in the workshop-style environment poses unique challenges due to class imbalance, fused action boundaries, and context dependence. This study proposes a multi-stage fine-tuning supervised framework for workshop-style HAR based on a 3D CNN, supplemented by an unsupervised post-assessment. A total of 208 clips (100 frames per clip) from two internal and two external datasets were used, along with 183 clips generated from 152 videos in a subset of the public UCF101 dataset for an add-on supplementary analysis. An I3D-ResNet50 (I3D_R50) backbone was used to examine a pilot model for domain adaptation and subsequently optimized for cross-context generalization. In Study 1, the I3D_R50 model with a customized 57-class prediction head was trained on the internal dataset, achieving a solid class- and cluster-level performance, however, spurious co-activations and limited separation among similar actions were observed. Study 2 refined sample allocation and data augmentation, and applied threshold tuning to evaluate within-context adaptation (Test 1) and cross-context generalization (Test 2). Multi-level analyses (class-, clip-, cluster-, and person-level) revealed moderate co-activation and semantic fusion within context, whereas cross-context generalization suffered from structure over-merging and semantic drift. As a post-assessment, Study 3 provided supplementary evidence that the model captured certain semantic co-occurrence relationships, although its performance remained constrained by limited IGE-TREE specific training samples and fine-grained behavioral categories. Limitations include insufficient samples, blurred temporal sequence, and context dependency. Future work could explore domain-adversarial alignment, pose-guided normalization, and temporal contrastive modeling, and selective merging of semantically similar actions to enhance semantic disentanglement and generalization.

The uncertainty and enormous potential of wave energy have drawn attention and research efforts on predicting offshore wave behavior to aid wave energy harvesting. The movement of offshore waves generates huge amounts of available renewable energy and creates a unique offshore energy source. Because offshore waves are mainly generated by wind, this paper focused on using wind speed as the main factor to predict offshore wave power density to assist wave energy harvesting. The dynamic behaviors of wave energy were displayed in this paper in a format of wave power density distribution, which was extracted and visualized in MATLAB. The model was reconstruction based on a long short-term memory (LSTM) neural network for one week and 3 h wave power density forecasting, integrated with wind conditions as input in two scenarios. One scenario explored the location effect for wave density forecasting. Another scenario compared the influence of different time series input of the structure. RMSE was used as a criteria estimator of the accuracy. The data period ranges from 1979 to 2019 in the Gulf of Mexico exacted from WaveWatch III. The lowest RMSE among different locations is 0.104, while the different time step scenario has an RMSE of 0.715. Because wind speed data is much easier to get from either hindcast dataset or actual measurement, the proposed method with the resulting accuracy will make the forecasting of wave power density much easier. The method has the ability to be implemented in other wave thriving locations, which fills the gap of forecasting on wave height and period based on buoy data given a lack of measurements, as well as reflecting the correlations between wind speed and wave density, thus providing support for a quantitative correlation model based on a deep-learning-based model.