為克服傳統電阻率成像因為耗時較長與容易受到人為影響而不易自動化的問題，有人提出以深度神經網路的方式為基礎的成像電阻率成像套件，此套件可以訓練好的神經網路，大幅縮短電阻率成像的時間。雖然訓練好的神經網路成像快速，但訓練過程卻非常耗時，也可能消耗許久時間訓練後跑出來的結果不盡如人意。再度生成訓練資料與改動參數重新訓練流程又是個複雜的手續，也無法知道程式完成時間點，預測的結果呈現也要直接至具有圖形處理器(Graphics Processing Unit， GPU)設備上才得知。為了克服上述問題，我們將使用到圖形處理器計算的程式改寫成應用程式介面(Application Programming Interface)，架設於有圖形處理器設備的裝置上作為雲端伺服器，再設計出一個圖形使用者介面(Graphical User Interface， GUI)客戶端應用程式負責傳入所需的參數。如此一來，只要在有網路的情況下，我們隨時隨地打開客戶端程式，便可知道訓練模型的進度，也隨時隨地可以停止或修改訓練資料與參數重新訓練或預測，更可以隨時觀看以前的預測結果，加上圖形化使用者介面，可以使這項技術普及率大幅提高。;In order to overcome the problem that the traditional resistivity images is challenging to be automation due to the long time consuming and anthropogenic influence, there is a proposal that package of resistivity images based on the Deep Neural Network can result in the positive neural network, which significantly shortens the time of the resistivity images. The positive neural network could be images quickly. However, the whole training is lengthy, and the result might be disappointing; even though renewing the process by generating the new training data and changing parameters is not only complicated but also not knowing the finish time, and the predicted outcomes have to be revealed until on the device, which has the function of the Graphics Processing Unit (GPU). For the purpose of solving the problem as mentioned above, we send the parameters to the Client by rewriting the calculated program with GPU into the Application Programming Interface (API) and setting it up on the device with GPU and then designs an application by Graphical User Interface (GUI) to the Client. As long as there is a network environment, we can receive the data of the training model and previous results, and even to stop or modify the training data and resend the parameters to the application for forecasting from time to time. With the GUI, the technology could increase its penetration of the market sharply.
