應用機器學習與深度學習於金屬零件影像識別之研究
|
摘要 本研究針對金屬零件的影像識別,探討了使用Teachable Machine和Matlab Transfer Learning App兩種方法實現AI模型自動識別的能力。在Teachable Machine平台上,成功建立了一個識別10種金屬零件的AI模型,並通過增加訓練照片數量的方式提高了模型的正確識別率。在Matlab Transfer Learning App中,使用9個預訓練的CNN卷積神經網路模型進行了識別實驗,結果顯示ResNet101模型的識別率最高,其次是Xception模型,而NASNetMobile模型表現最差。本研究的結果顯示,Teachable Machine是一個方便易用的平台,可以快速構建AI模型;遷移學習是一個快速且有效的方法,可以利用預訓練的模型來加快新模型的訓練過程。使用者可以根據自己的需求選擇最適合的方法和模型。
關鍵字:金屬零件、影像識別、Teachable Machine、Matlab Transfer Learning、CNN卷積神經網路模型。 ABSTRACT This study investigates the ability to use two methods, Teachable Machine, and Matlab Transfer Learning App, to implement AI models to automatically recognize metal parts in images. A successful AI model was built on the Teachable Machine platform that can identify 10 types of metal parts, and the accuracy of the model was improved by increasing the number of training photos. In Matlab Transfer Learning App, experiments were performed using 9 pre-trained CNN convolutional neural network models for the recognition, and the results showed that the ResNet101 model had the highest recognition rate, followed by the Xception model, while the NASNetMobile model performed the worst. The results of this study demonstrate that Teachable Machine is a convenient and user-friendly platform for quickly building AI models, and transfer learning is a fast and effective method that can use pre-trained models to accelerate the training process of new models. Users can choose the most suitable method and model according to their own needs. KEYWORDS: Metal Part; Image Recognition; Teachable Machine; Transfer Learning; CNN Convolutional Neural Network Model. |