張參數：印刷電路板瑕疵檢測演算法參數推薦

Abstract

印刷電路板（PCB）元件檢測對於電子產業中的品質和功能保證至關重要。本研究旨在開發一個自動化的PCB引腳缺陷檢測算法參數推薦系統，重點在於準確識別引腳元件資料中的heel和toe位置。傳統的算法參數調整方法高度依賴產線工程師的經驗，這種方法既不一致又耗時。為了解決這個問題，我們提出了張參數演算法，結合了一維變分自編碼器（1D VAE）進行數據生成和一維殘差網絡（1D ResNet）進行模型訓練。 1D VAE被用來生成多樣化的數據，增強訓練數據集並提升模型的泛化能力。隨後，1D ResNet在原始數據和生成數據上進行訓練，以準確預測heel和toe的位置。我們進行了綜合實驗來評估所提出算法在不同PCB資料集上的性能。結果顯示，張參數演算法能有效提升模型的泛化能力，適應各種PCB配置，提供可靠的參數推薦。

The inspection of Printed Circuit Board (PCB) components is critical in the electronics industry to ensure the quality and functionality of electronic devices. This study aims to develop an automated parameter recommendation system for PCB pin defect inspection algorithms, focusing on accurately identifying heel and toe positions in pin component data. Traditional methods for adjusting algorithm parameters rely heavily on the experience of production line engineers, which can be inconsistent and time-consuming. To address this, we propose ChangParameter algorithm, which uses a combination of a one-dimensional Variational Auto-Encoder (1D VAE) for data generation and a one-dimensional Residual Network (1D ResNet) for model training. The 1D VAE is employed to generate diverse data, enhancing the training dataset and improving the model's generalization capabilities. The 1D ResNet is then trained on both the original and generated data to predict the heel and toe positions accurately. We conduct comprehensive experiments to evaluate the performance of the proposed algorithm across different PCB datasets. The results demonstrate that ChangParameter algorithm effectively enhances the model's ability to generalize and adapt to various PCB configurations, providing reliable parameter recommendations.