Search for: All records

Annotated IMU sensor data from smart devices and wearables are essential for developing supervised models for fine-grained human activity recognition, albeit generating sufficient annotated data for diverse human activities under different environments is challenging. Existing approaches primarily use human-in-the-loop based techniques, including active learning; however, they are tedious, costly, and time-consuming. Leveraging the availability of acoustic data from embedded microphones over the data collection devices, in this paper, we propose LASO, a multimodal approach for automated data annotation from acoustic and locomotive information. LASO works over the edge device itself, ensuring that only the annotated IMU data is collected, discarding the acoustic data from the device itself, hence preserving the audio-privacy of the user. In the absence of any pre-existing labeling information, such an auto-annotation is challenging as the IMU data needs to be sessionized for different time-scaled activities in a completely unsupervised manner. We use a change-point detection technique while synchronizing the locomotive information from the IMU data with the acoustic data, and then use pre-trained audio-based activity recognition models for labeling the IMU data while handling the acoustic noises. LASO efficiently annotates IMU data, without any explicit human intervention, with a mean accuracy of 0.93 ($\pm 0.04$) and 0.78 ($\pm 0.05$) for two different real-life datasets from workshop and kitchen environments, respectively. 

We mapped tidal wetland gross primary production (GPP) with unprecedented detail for multiple wetland types across the continental United States (CONUS) at 16‐day intervals for the years 2000–2019. To accomplish this task, we developed the spatially explicit Blue Carbon (BC) model, which combined tidal wetland cover and field‐based eddy covariance tower data into a single Bayesian framework, and used a super computer network and remote sensing imagery (Moderate Resolution Imaging Spectroradiometer Enhanced Vegetation Index). We found a strong fit between the BC model and eddy covariance data from 10 different towers (r²= 0.83,p< 0.001, root‐mean‐square error = 1.22 g C/m²/day, average error was 7% with a mean bias of nearly zero). When compared with NASA's MOD17 GPP product, which uses a generalized terrestrial algorithm, the BC model reduced error by approximately half (MOD17 hadr²= 0.45,p< 0.001, root‐mean‐square error of 3.38 g C/m²/day, average error of 15%). The BC model also included mixed pixels in areas not covered by MOD17, which comprised approximately 16.8% of CONUS tidal wetland GPP. Results showed that across CONUS between 2000 and 2019, the average daily GPP per m²was 4.32 ± 2.45 g C/m²/day. The total annual GPP for the CONUS was 39.65 ± 0.89 Tg C/year. GPP for the Gulf Coast was nearly double that of the Atlantic and Pacific Coasts combined. Louisiana alone accounted for 15.78 ± 0.75 Tg C/year, with its Atchafalaya/Vermillion Bay basin at 4.72 ± 0.14 Tg C/year. The BC model provides a robust platform for integrating data from disparate sources and exploring regional trends in GPP across tidal wetlands.