Research On Image Content Automatic Classification Based On ShuffleNet Network Model

With the rapid development of software and hardware technology for android mobile phone photographing,users have become more dependent on the photographing function.Whether it is recorded about the moments of happiness in life,such as gourmet and photographing,or capturing the meeting records in the work and recording the project progress,the frequency of camera use has gradually been increased.And in addition to photographs,the user will save some important images for studying,work and so on.As a result,the increase in the number of images makes it difficult for the user to quickly find the target image from thousands or even tens of thousands of images.The traditional way of manually searching for mobile phone images is time-consuming,and most of the existing images organization management applications listed on the market are classified.In view of the above problems,there is a need for a more intelligent and efficient automatic image classification method.To this end,the main research contents are as follows:(1)Images organization and management carried out from the date,position and images content into three perspective.That is: 1)First,the phone images are sorted according to the photographing time or the storage time.2)Based on the time-based image classification,the mobile phone images are categorized according to the position information.3)For the time-position categorized images,they are further automatically classified according to the content.The classification categories include: food,night scene,PPT,blackboard,self-timer,clothes,shoes and other categories.(2)Implement the above-mentioned multi-level organization management method based on ShuffleNet network model.In this paper,ShuffleNet lightweight convolutional neural network structure is built under the TensorFlow platform.Training and validation analysis were performed using ReLU and Leaky ReLU activation functions,and a ShuffleNet network was built using the Leaky ReLU activation function.Based transfer learning,migration technology will be applied to the multi-level network ShuffleNet image organization and management methods,reaching 97.4% accuracy.The experimental results from the acquired data sets show: Based on the time,position information and content of the image organization and management methods,the search process can be gradually reduced in the search process,enabling users to find a wide range of target image in just a few seconds,greatly reducing the search time,saving time and effort.