發展液晶數字儀表之智慧化自動判讀系統

台灣自來水公司因應人工智慧管理趨勢，且為了達到節約能源、精簡人力及避免人力斷層，積極整合取水、產水、輸水等監控管理系統，規劃逐步朝向視覺化及網頁化之監控整合管理系統發展。鑒於台灣自來水公司使用之設備大多為液晶數字儀表板，且尚未全部納入可程式控制器的監控範疇，若能搭配視覺影像資料，與監控資訊分析比對，能於出現異常數值時，即時確認正確資訊，為此，本文提出發展一套液晶數字儀表之智慧化自動判讀系統，以結合機器視覺、機器學習、卷積神經網路及影像識別分析等功能，與液晶數字儀表顯示之監控數值交叉比對，提供管理人員在遠端即可獲知正確訊息，縮短問題處理時間並提升效率。本文利用影像處理、深度學習之卷積神經網路架構，以液晶型式的阿拉伯數字進行訓練，再判讀及辨識液晶顯示器上的數據。根據本文的實驗結果顯示，正確率可達98%以上。

關鍵字

深度學習；機器學習；機器視覺；影像識別；卷積神經網路

並列摘要

To adjust the artificial intelligence management tendency , save energy and downsizing manpower to avoid manpower fault, Taiwan Water Supply Company actively integrates monitoring manage system for getting , product and delivery, water.Gradually develop toward a visual ,web-based monitoring manage system. Since most of Taiwan Water Supply Company equipments use LCD digital instrument panel, not included all of the monitoring category of programmable controllers. If it can be arranged with visual image data and compared with monitoring information analysis then will be alert in real time when abnormal occurred. For this purpose, this article proposes to develop a set of intelligent automatic interpretation system for liquid crystal digital instruments.The system combines such as machine vision, learning, convolutional neural network, image recognition analysis functions and also can cross-compare with the monitored values displayed by liquid crystal digital instruments.Can provide managers to get the correct information remotely and shortening the problem processing time to improve efficiency.This article uses the convolutional neural network architecture of image processing and deep learning to train with liquid crystal Arabic numerals, then interpret and recognize the data on the liquid crystal display. According to this article experiment results, the correct rate can reach more than 98%.

並列關鍵字

deep learning ； machine learning ； machine vision ； image recognition ； convolutional neural network.

參考文獻

[1] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” in Proceedings of the IEEE, Vol. 86, pp. 2278-2324, 1998.