焊點是電子元件和電路板的相會之處，良好的焊接可以讓電路正常運作，然而有瑕疵的焊接會讓整個電路產生不可預期的錯誤，因此焊點的品質對產品的成敗有直接的關係。 過去的自動化視覺檢測系統仰賴人所制定的規則(rule-based)，其修正過程充滿不確定性，本研究通過深度學習以訓練智慧型機器視覺系統，使其能夠辨識焊點的好壞。 Xception是google繼Inception架構而生的神經網路架構，本論文使用Xception對焊點進行訓練。利用合格與品質不良的焊點樣本經過卷積之後所產生的特徵圖之差異訓練智慧型機器視覺系統，使該系統能夠分辨出焊點之良莠。 ;Solder joints are the intersection of electronic components and circuit boards. Good soldering allows the circuit to operate normally. However, flawed soldering can cause unpredictable errors in the entire circuit. Therefore, the quality of solder joints has a direct impact on the quality of the electronic product. In the past, the automated visual inspection system usually functions in rule-based fashion, and the fine-tuning process was full of uncertainty. In this study we apply deep learning paradigm to train the neural network model to identify the quality of the solder joints. Xception is a neural network architecture which inherits the concept of Inception created by Google. This paper uses solder joints to train Xception. A neural network model trained with the difference between the feature maps produced by the convolution of Pass and Ng solder joint samples can identify the quality of the solder joints.
