本研究針對家長在課後教導小孩在書寫品質評估上之人力負擔，提出一套邊緣智慧與即時通訊技術結合之互動式書寫輔助系統。系統以YOLO偵測作業簿中的手寫字，並以CNN進行單字辨識；同時採用深度特徵空間（Deep Feature Space）進行分析，並結合 Prototype-based representation 與多指標融合機制作為字形相似度評分，於終端裝置（Jetson Orin Nano）側完成即時推論，根據結果提供評分與評語。實驗在自建資料集（含 4803 常用字標準字庫與多來源手寫樣本）上進行，評估檢測 mAP、分類 Top-1/Top-5。結果顯示，本系統在無需家長親身在旁的前提下，能穩定完成手寫字擷取、辨識與評分。 本研究之貢獻包含：（1）在邊緣端完成偵測/分類/評分之整合；（2）建立可重現之資料集與評估流程，供後續研究與實務導入。;This study proposes an interactive handwriting assistance system that integrates edge intelligence and real-time communication technology to reduce the burden on parents when evaluating children’s handwriting quality after class. The proposed system employs YOLO to detect handwritten characters in workbooks and utilizes a Convolutional Neural Network (CNN) for character recognition. In addition， a deep feature space analysis method is adopted， combining Prototype-based representation and a multi-metric fusion mechanism for handwriting similarity scoring. Real-time inference is performed on an edge device， namely the Jetson Orin Nano， to generate evaluation scores and feedback comments based on the recognition results. Experiments were conducted on a self-constructed dataset consisting of 4803 commonly used Chinese character templates and handwritten samples collected from multiple sources. The evaluation metrics included detection mAP and classification Top-1/Top-5 accuracy. Experimental results demonstrate that the proposed system can stably accomplish handwritten character extraction， recognition， and scoring without requiring continuous parental supervision. The main contributions of this study are as follows: (1) The integration of detection， classification， and scoring modules on an edge-computing platform. (2) The establishment of a reproducible dataset and evaluation framework for future research and practical applications.
