Human activity recognition using sensor data has become a research hotspot in the field of ubiquitous computing and has a wide range of application scenarios in real life. Effectively extracting and fusing the rich spatiotemporal information contained in the sensor data from various sensors and multiple wearable devices is a challenging task. Moreover, for practical deployment applications, the limited computational and storage resources of edge devices also impose requirements on the model size and inference speed. Hence, we propose STFNet, an enhanced and lightweight spatiotemporal fusion network for wearable human activity recognition. In which, we innovatively divide the extraction and fusion process of spatiotemporal features into three stages, including sensing feature extraction, temporal feature extraction, and channel fusion, to learn incrementally and reduce information loss. We first extract shallow intra-channel features and learn local inter-channel information by using a lightweight local attention mechanism. Then, the learned features are fed into a hierarchical convolution structure with residual connections to extract multi-scale and comprehensive temporal features. Finally, we use graph convolutions to further fuse the deep inter-channel information. By replacing the conventional global attention mechanism with a local one and using hierarchical convolution to replace dilated or multi-kernel convolution, STFNet achieves the purpose of lightweight design. In the experiments, STFNet performs better on public datasets than the selected WHAR models published in recent years and demonstrates its lightweight characteristics on edge device deployment.

The continual progression and expanding applications of hyperspectral image (HSI) technology necessitate the development of lightweight HSI classification models that are capable of lifelong learning. However, the computationally-demanding task of training and updating HSI classification models, exacerbated by the substantial number of trainable parameters in feature extractors, remains a substantial challenge. This article proposes an attention-based reservoir computing (ARC) model to overcome these hurdles. The ARC model utilizes a cross-slicing operation to generate multidirectional inputs, treating the HSI dataset as a spatial sequence for processing within a reservoir. For every target pixel, four spatial sequences from various directions are introduced into the reservoir, generating four corresponding outputs. A voting mechanism then evaluates these outputs to yield the final prediction. In addition, we design an attention-based leaky function for reservoir computing (RC) to capture the spatial correlation inherent in HSI data accurately. The attention-based leaky function enables the reservoir state to weigh less on the pixels outside the region of interest (ROI) and have a longer memory for pixels inside the ROI. The ARC was tested on widely used HSI datasets: Indian Pines, PaviaU, and Salinas. It demonstrated competitive lightweight classification performance against state-of-the-art lightweight models by maintaining comparable training time while achieving superior accuracy. Furthermore, the model’s lifelong learning accuracy also showed superior performance compared to existing lifelong learning models, with a one thousand times reduction of parameter-to-be-updated. This work makes the ARC model an effective contender for HSI classification tasks, excelling in both lightweight classification and lifelong learning capacities. The source codes are available publicly at: https://github.com/Waterman-Ann/ARC .

Harmful Algal Blooms (HAB), a significant environmental concern on a global scale, are damaging to ecosystem functions and pose challenges to environmental management. The key to HAB management is to classify the local representative species. However, the widespread adoption of automated monitoring devices such as the Imaging FlowCytobot (IFCB) is cost-prohibitive and often cannot be efficiently utilized. An algae classification algorithm based on transfer learning designed to classify local algal species with limited low-cost microscope-based algae images is developed to offer a cost-effective and practical solution for HAB management. The model is tested on the Macau representative freshwater algae dataset. The training process of the transfer learning model is based on the automatically captured open-source dataset and IFCB-based Hong Kong tropical algal dataset. The results demonstrated the exceptional performance of our method, achieving up to 100% accuracy in the classification of single algal species. Furthermore, when compared with commonly used deep learning models in experimental comparisons, our method exhibited superior performance. We also conducted ablation experiments on transfer learning, which validated the effectiveness of our transfer learning approach.

With the increasing number of Hyperspectral Image (HSI) classification scenarios and frequent updates to classification categories, there is a growing demand for new models and updates to existing ones. In this paper, we develop a lightweight lifelong HSI classification model called Attention-based Reservoir Computing (ARC). This model eliminates the need for training and updating the feature extractor, largely reducing computational overhead and enabling fast adaptation to new classes. The proposed ARC’s superior performance, lightweight nature, and lifelong learning abilities promise to enhance the accuracy, efficiency, and adaptability of HSI classification, contributing to improved natural resource management, environmental monitoring, and disaster warning systems.