Research on optimization technology based on mobile terminals convolutional neural networks

Abstract

Convolutional neural networks play a very important role in computer vision, such as image classification, image segmentation, and handwriting recognition have been widely used. In daily life, this technology is used in the photo recognition of e-commerce platforms. However, the timing of the identification process grows into a major problem. Therefore, it is particularly important to reduce the recognition time by optimizing the deep learning model. To solve this problem, two experimental methods are proposed to optimize the volume of the convolutional neural network model. The first is to reduce the size of the model by scaling down the convolutional kernel. The second is to prune the model with L-1 norm to reduce the size of the model and improve the running speed. According to the experimental results, the two experimental methods have achieved remarkable optimization effects. In the first experiment, the method of scaling down convolutional kernel has an important optimization effect for training the deep learning model of small data sets. In another experiment using L-1 pruning algorithm greatly improves the running speed of models by reducing the size of models. To sum up, the optimization method proposed above for the convolutional neural network model on the mobile end can be applied in the field that requires a large amount of image classification, such as delivery package sorting. At the same time, to better improve the performance of the model, it will become feasible to use a variety of optimization methods to tune it.