近年來基於深度學習方法的影像辨識相關應用需求不斷增加，對於開發者的負擔也隨之倍增，因此本論文設計一個具有Low-code性質的影像辨識平台來達到快速開發的目的，並且使用2021年新推出的神經網路模型-MLP-Mixer來做為本系統的神經網路架構。本研究開發了一個圖形化人機介面讓使用者能快速地訓練及測試神經網路模型，並使用三種不同的影像數據集進行實驗與分析，準確率分別達到85%、96.5%及89.6%，也驗整了本平台能夠實現在不同數據集上的影像辨識應用。本論文所提出的MLP-Mixer影像辨識低代碼開發平台，在進行訓練、測試模型和分類預測的全部過程中，僅需要選取資料夾和輸入相關參數即可自動完成，此Low-code的特性讓非專家的一般使用者也能輕鬆地操作。;In recent years, the demand for image recognition related applications based on deep learning methods has continued to increase, and the burden on developers has also doubled. Therefore, this paper designs a low-code image recognition platform to achieve rapid development purposes, and the MLP-Mixer which is the newly launched neural network model in 2021 is used as the neural network architecture of the system. This research has developed a graphical human-machine interface that allows users to quickly train and test neural network models, and uses three different image datasets for experiments and analysis, with accuracy rates of 85%, 96.5%, and 89.6%, respectively. It has also been verified that the platform can realize image recognition applications on different datasets. The MLP-Mixer image recognition low-code development platform proposed in this paper can be automatically completed by selecting the folder and inputting relevant parameters in the entire process of training, testing the model and classification prediction. This low-code feature allows Non-expert general users can also easily use.
