Research On Image Caption Based On Mobile Computing Environment

Image Caption has a huge amount of value for applying on mobile computing environment,which can bring new functions to mobile application software and improve user experience.The traditional way of applying of deep learning on mobile terminals not only requires manual compilation of models,but also considers the stability of mobile devices' network connection and the size of storage.For complex neural network models,such as image caption models,it often takes more storage,while simplifying the model will cause the accuracy of the model to decrease.In addition,due to the limitation on the speed of mobile devices,the complex neural network model is difficult to run efficiently on mobile devices.To solve these problems,this paper did a research on the method of applying image caption on mobile devices in consideration of mobile computing environment.The main contents include:1.Take into account the instability of network connection of mobile devices,this paper proposes running a neural network model of image caption directly on the mobile devices.Mobile devices can still do image caption while the mobile devices having no network connection,moreover the response time can be reduced,which lead to the real-time performance improved.2.For running a neural network model on mobile devices always causing a decrease in accuracy and low efficiency,this paper proposes using Core ML,which is a framework designed to develop neural network models on mobile devices.Through this method,developers can ensure the accuracy of the model and the computation speed,when applying image caption in mobile devices.In this paper,the application of image caption in iOS mobile devices is successfully implemented in the experiment.The comparative test results show that: through the optimization of Core ML on the mobile computing model,on mobile devices running accurately and fast,which had an excellent performance close to the performance on PC.Thus,this paper proved that applying image caption on mobile devices by using Core ML is superior to the traditional way.