LEGO Minifigure Recognition Based on Deep Learing

LEGO® is a building block company that allows children and adults to create endlessly changing shapes through brainstorming. It is so admired that people call it “Magic Plastic Building Blocks”. Exquisite LEGO Minifigure of various themes also increase the fun of Lego bricks. This paper uses deep learning methods to predict and recognize LEGO Minifigures. This paper employs four different convolutional neural network models for training. Due to the tiny dataset size, the experimental results were suboptimal. To address this issue, transfer learning is introduced to mitigate the overfitting problem. Additionally, three different optimizers are utilized to conduct the final evaluation of the model's performance. After applying the transfer learning method, the experimental results show that using the VGG 16 model with an RMSProp optimizer has the best performance.