Abstract
ABSTRACTOptical character recognition (OCR) based on wearable devices plays an important role in online learning. Although the existing convolutional recurrent neural network (CRNN) has achieved great success in the task of optical character recognition (OCR), this model cannot fully achieve global contextual information modeling. With the development of artificial intelligence IoT (AIoT) technology, existing deep models are difficult to deploy on resource‐constrained mobile devices. Therefore, this article proposes an effective context‐aware TinyML model for Korean handwriting recognition. Specifically, we effectively improve the global context modeling capability by embedding the Mamba layer in CRNN with linear computational complexity. In addition, we achieved model compression by introducing a distillation mechanism based on a multi‐layer joint distillation mechanism. A large number of experimental results on two publicly available datasets of Korean characters show that our proposed model achieves a higher performance.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call