近年來,深度學習技術被大量應用在多個不同的領域,其中虛擬實境頭盔結合深度學習技術常被使用來提升生產效率。但是因為在訓練深度學習模型時會需要大量的訓練資料,而工廠的作業流程通常都是在接到訂單之後才會開始生產,因此很難預先取得大量的訓練資料供深度學習模型來訓練,故使用 3D 零件模型影像來訓練模型及辨識真實零件影像便是一個較為可行的解決方式。雖然使用 3D 零件模型影像來訓練模型可以解決缺乏訓練資料的問題,但是工廠提供的 3D 零件模型中並不一定會包含紋理資訊,導致其和真實零件還是有很大的差距,因此若要使用這些 3D 零件模型影像來訓練模型辨識真實零件會是一項困難的挑戰。而在生成的 3D 零件模型影像中,部分角度之零件影像會容易使模型混淆,且因為零件之間具有很高的相似性,所以導致模型會有預測準確度不佳的問題,因此本研究提出了依據零件面積比例來過濾資料集,去除掉較容易使模型混淆之影像,此外,本研究也提出了長度過濾模組來輔助模型推論,通過長度資訊篩選掉較不符合之類別,實驗結果顯示,我們提出的方法可以顯著提升模型在細粒度真實工業零件分類問題的表現。;In recent years, deep learning technology has been widely used in various fields. One com mon application is combining virtual reality helmets with deep learning to improve production efficiency. However, training deep learning models requires a lot of data, which is difficult to obtain in factories where production starts only after orders are received. Using 3D models of components to train the model and recognize real component images is a feasible solution, but the lack of texture information in 3D models provided by the factory poses a significant chal lenge in accurately identifying real components. To address this, our study proposes a filtering method based on component area ratios to eliminate confusing images, and introduces a length filter module to assist model inference by filtering out mismatching size categories. Experimen tal results show that our methods significantly improve model performance in fine-grained real world industrial component classification tasks.