建立H型鋼構件的全自動化塗裝產線的第一步便是建立辨識系統。因此，本研究目的在於開發運用深度神經網絡的H型鋼構件自動辨識系統。透過文獻回顧，深度神經網路在辨識方面有著良好的表現。其為所提出的演算法提供核心計算能力，並根據其他8個步驟包括數據的輸入，偵測平面之計算與剔除，對剩餘平面進行分類，計算平行平面可能的間距，確定最佳辨識的構件資訊，對實際和辨識的構件進行比較，輸出至Matplotlib軟件使用所輸出的空間座標資訊建置三維模型，以及完成詳細的三維表面模型建置。透過從塗裝工廠隨機獲得的115件模擬案例和99件實際案例進行驗證， H型鋼構件整體的辨識準確率均高於99.12％，其中大部分的辨識率達到100％，而這些部分包括H型鋼構件的寬度，腹板厚度和翼緣板厚度。本研究的結果表明所提出的系統是可靠的，其平均準確率為99.73％，並且可被實際應用。;The first step to establish a fully automatic coating production line for H-steel components is to set up a recognition system. Therefore, the research purpose is to develop a recognition system for H-steel components using Deep Neural Networks (DNN). Literature review suggests that DNN performs well in recognition. It gives the proposed algorithm the core computation, followed by the other 8 steps of data input, calculation for detected area, classification for residual area, possible distances among parallels, determination for optimal recognition, comparison between actual and recognized components, output to Matplotlib software, 3D plot using output coordinates, and completion of detailed 3D plot. The evaluation is carried out using 115 simulated and 99 actual cases randomly obtained from the factories. The accuracy rates for all parts of H-steel components are higher than 99.12% where 100% recognition rate is reached for most parts such as H-steel width, base plate thickness, and wing plate thickness. The findings support that the proposed system is reliable with an average accuracy rate at 99.73% and applicable in reality.
