本實驗的蝴蝶圖像取自ImageNet，共8500張圖片，並自製成數據樣本集，將訓練集分別帶入上述模型後，觀察個別訓練時間及訓練準確率之差異，並在迭代結果上進行比較。而後再進一步探討影響訓練結果的原因。最後將測試集放入訓練好的模型進行預測，觀察測試集準確率，分析探討影響預測結果的因素。 ;The goal of this thesis is to explore the training results of “K Nearest Neighbor”, “multilayer perceptual neural network” , “Support Vector Machine” and the classic model of Convolutional neural network: “LENET” and “ALEXNET” in image recognition.
The butterfly images in this experiment are from ImageNet which is the largest database of image recognition. First, we bring the training data into our models, and observe the difference between training time and training accuracy for each model, then compare the iterative results. Next,we give the reasons that affect the training results. Finally, we put the test set into the trained model for prediction.We observe the accuracy of the test set, and analyzed the factors affecting the prediction.
