Deep Transfer Learning for Detection of Upper and Lower Body Movements: Transformer With Convolutional Neural Network
Bibliographic record
Abstract
When humans repeat the same motion, the tendons, muscles, and nerves can be damaged, causing repetitive stress injuries (RSIs). If the repetitive motions that lead to RSI are recognized early, actions can be taken to prevent these injuries. As human activity recognition (HAR) aims to identify activities employing wearable or environment sensors, HAR is the first step toward identifying repetitive motions. Deep learning models, such as convolutional neural networks (CNNs), have seen great success in recognizing activities for participants whose data are used in the model training; however, their accuracy drops for new participants as people move in different ways. Moreover, most studies focus on lower body movement, while upper body movements are the main cause of RSI. On the other hand, in recent years, transformers have been dominating natural language processing (NLP) and have the potential to improve modeling in other domains involving sequential data such as HAR. Consequently, this article combines a transformer and CNN (Trans-CNN) for the recognition of upper and lower body movements. Transfer learning was employed to personalize the generic model for the target participant. The experiments demonstrate that the generic Trans-CNN outperforms the standalone Trans-CNN. The accuracy of the generic Trans-CNN for both upper and lower body movements improved from 69.6% to 92.4% when personalization was introduced. All models, irrespective of the algorithm, have more difficulty recognizing the upper body than lower body movements. Nevertheless, the proposed personalized approach for the detection of upper and lower body movements represents significant progress toward RSI prevention.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".