GAF-ConvNeXt Algorithm With Transfer Learning for Human Activity Recognition Based on Wearable Plastic Fiber Sensors

Human activity recognition (HAR) is an important research topic in the field of human–computer interaction (HCI), and wearable HAR based on convolutional neural networks (CNNs) has achieved the most advanced performance. In this study, wearable plastic optical fiber (POF) sensors based on the GAF-ConvNeXt algorithm are used for HAR for the first time. The GAF-ConvNeXt algorithm, combined with the Gramian angular field (GAF) and the latest CNN model ConvNeXt, can convert motion signal of a 1-D time series into a 2-D image, which can not only retain the complete information of the signal, but also maintain its dependence on time, thereby presenting more depth features and facilitating accurate recognition of human activities. Furthermore, transfer learning (TL) was introduced to further optimize the GAF-ConvNeXt model, which effectively reduces the number of training samples, improves the recognition accuracy, and accelerates the convergence speed. The GAF-ConvNeXt algorithm based on TL can recognize seven motion modes with an accuracy of 99.75%, and has great application potential in the fields of human motion recognition and HCI.