Research And Application Of Target Detection Algorithm Based On Side-to End Cooperation

The deep target detection technology in computer vision is a very important technology.In order to accurately extract valuable information,it is very important to locate and classify objects in videos or images.However,there are some problems.The cloud-based deep learning model has a long response delay,and due to the instability of the mobile network and the limited network bandwidth,these will affect the user experience.At the same time,due to the limited resources of mobile devices,deep learning models have high requirements on the computing power and storage capabilities of the devices,and cannot be directly deployed on mobile devices with limited resources.Therefore,how to deploy deep learning tasks to the edge computing environment is a problem worth studying.In order to solve the above problems and meet the requirements of safe,accurate and efficient image data processing,this paper combines deep learning technology and edge computing technology to conduct the following research:(1)From the perspective of target detection research,considering the limited computing resources of the mobile terminal,the in-depth model requires large computing power,the communication between the mobile terminal and the edge terminal,etc.,the real-time depth is formulated by using the real-time performance of the mobile terminal and the computing power of the edge terminal.The deployment strategy of target detection ensures the smooth progress of target detection tasks.The Oxford flower data set was used to simulate the mobile image classification model.Within the acceptable accuracy range,the model size can be reduced to 2/11.The public data set VOC2007 data set is used to verify the edge target detection model YOLOv5.The experimental results show that its R value is0.965 and m AP@0.5 is 0.964,which has great advantages compared with YOLOv3.(2)In order to reduce the amount of data that needs to be transmitted between the mobile terminal and the edge terminal,this paper designs partial image compression,and uses partial compression for different areas to reduce the amount of data transmission.Use the public data set VOC2007 to carry on the simulation experiment to part of the image compression part.When the compression rate of the background area is set to 60%,the image data size is reduced by 8/9.(3)Aiming at the problem that it is difficult to collect panoramic images in some specific application scenarios,combined with some methods of digital image processing,an image stitching and fusion strategy based on edge computing is implemented.The experimental results of the image splicing and fusion strategy show that the strategy can splice and fuse multiple pictures without stitching.