A comparative research on clothing images classification based on neural network models

To make deep learning really work in real life, it is better for the model to be used with neural networks of small memory requirements and small computations. In this paper, four neural network models, i.e. Fully Connected Neural Network, CNN, MobileNetV1, and MobileNetV2 are applied to deal with the classification of clothing images in the Fashion-MNIST dataset and the results are compared. MobileNetV2, which performs best, is a deep learning network in the field of image classification and recognition. With the development of robots, smart auto, and augmented reality in real-world application, it is better to use a network such as MobileNet in a computationally limited platform. The experimental results show that 1) MobileNet is a more effective method of classification of clothing images compared to Fully Connected Neural Network and CNN, with an accuracy rate of 91%. 2) MobileNetV2 has greater improvement in accuracy than the previous model, which accuracy has reached to 93%. 3) MobileNet is simpler than Fully Connected Neural Network and CNN model structures, although training time has increased, it ensures higher accuracy.