Research And Implementation Of Distributed Object Detection Framework For Mobile Environment

With the continuous development of mobile internet and mobile devices,artificial intelligence is widely used in mobile environment.Object detection is a basic computer vision task.Object detection technology have produced a wealth of intelligent applications on mobile devices,and mobile environment has also brought data and efficiency problems to object detection:(1)mobile environment makes the application of artificial intelligence more extensive,but also puts higher requirements for model accuracy,and puts higher requirements for data breadth,In order to meet the needs of mobile environment,we need to expand the data constantly;(2)there are more images,more complex deep learning model and more abundant application programs in the mobile environment,which put higher requirements for the computing efficiency of the mobile terminal.These two problems must be solved first,in order to achieve the widespread application of object detection models in mobile environments.The goal of this paper is to design and implement distributed object detection for mobile devices,improve the ability of object detection task processing for mobile terminals.the main research contents of this paper are as follows:(1)A training algorithm of object detection model for decentralized data is proposed,which achieves multi-party cooperative and effective training of object detection model,optimizing the object detection model through federated learning algorithm and alleviating the bad influence of the unbalanced distribution of decentralized data;(2)Intelligent collaborative inference method of the object detection model based on reinforcement learning is proposed,which realizes the automatic task allocation and node placement of the object detection model in mobile devices,and have the ability to adjust the allocation scheme according to the mobile environment,effectively improve the speed of the model inference.Finally,the implement of intelligent distributed object detection model based on this method increase the speed by 10.6%on multiple heterogeneous mobile devices;(3)The intelligent and efficient distributed object detection for mobile environment established.These three contents study the effective application of data in mobile environment,the efficient reference scheme of object detection model on mobile devices and the construction of distributed object detection,and jointly complete the research and implementation of object detection framework for mobile devices.The research and implementation of distributed object detection framework for mobile environment can expand the available data and improve the efficiency of the model in mobile environment.More accurate and faster object detection algorithm can promote more computer vision applications to apply in mobile environment.