A Method for Optimizing Multi-Scale Learning Networks Based on MobileNets

Abstract

The present invention relates to a method for optimizing a multi-scale learning network based on MobileNets. The multi-scale learning network of the present invention includes 4 parts, the first 3 parts are the same separable convolutional layer, each separable convolutional layer is connected with batchnorm and ReLU, then connect the pooling layer, and finally connect with the fully connected layer and the output layer in part 4, where the separable convolution layer includes 3 sets of convolution operations, the specific network structure is, the first set is 3*3 depth convolution The convolution operation is performed by the product, and the second group uses two 3*3 depth convolutions for convolution operation, and then adds the output of the first group and the second group, and continues to use 1*1 point convolution convolution operation; the third group directly uses 1*1 point convolution to perform convolution operation, and then merges the outputs of the first, second and third groups; through experimental comparison, it is found that the network constructed by the present invention The structure experiment parameters are few, the precision is high, and the structure of the three separable convolutional layers is stable, and the experiment effect is the most ideal.

Description

technical field [0001] The invention belongs to the field of image classification, which is mainly applied to mobile and embedded vision applications, and is a lightweight deep neural network proposed for embedded devices such as mobile phones. Image classification is an image processing method that distinguishes different types of objects through the different characteristics reflected in the image information. It uses computer to carry out quantitative analysis on images, and classifies each pixel or area in the image or image into one of several categories to replace human visual interpretation. Background technique [0002] In the context of the development of deep learning, convolutional neural networks have been recognized by more and more people, and their applications are becoming more and more common. The current general development trend of deep learning is to obtain higher accuracy through deeper and more complex networks. However, these deeper and more complex n...

AI Technical Summary

Problems solved by technology

However, compared with other networks with the same number of layers, the existing MobileNets network has optimized the time and number of parameters, but the accuracy is not optimistic.

Images