Classification of Electronic Components Using MobileNetV2 Architecture-Based Convolutional Neural Network

Research on Convolutional Neural Networks (CNN) continues to be developed to increase efficiency and speed in processing data and reduce the need for large amounts of training data. The CNN network architecture used in this study is MobileNetV2. The MobileNetV2 architecture is used in the transfer learning process, which is a learning technique where the model that has been trained is used as a basis for learning other tasks, in this case, for image recognition. In the CNN algorithm training process, four layers are used in the electronic component image classification process. The layers used include flattened, dropout layer 0.5, dense 128 layers with Relu activation, and dense 15 layers with a softmax activation function. The image dataset used is 5.716 images divided into 15 different classes. The dataset that becomes the input data will be divided into training and test data. To produce the highest accuracy, the model will be developed parameters. The model parameters developed in this study are the dataset ratio, learning rate, and the number of epochs. The test results from the CNN MobileNetV2 model produce the highest accuracy value of 96.42 and the lowest loss value of 11.97. At this highest accuracy, the learning rate is 0.01 with 50 epochs and a dataset comparison of 80%:20%. In this 80%:20% dataset comparison, 143 batches are needed to complete one epoch.