Research And Application In Object Cognition System Based On Image Recognition Algorithm

Object cognition is a technology that uses the image,shape,sound,smell and other data collected by the sensor to identify objects.Object cognition system usually bases on embedded computers.Traditional object cognition systems base on sensing technology,such as RFID,can quickly track object and exchange data.However,RFID technology relies on electronic tags on objects,which has high cost and distance limitation.Images have gradually become a part of the Big Data of the IoT with the development of hardware equipment and network technology.At the same time,the image recognition technology based on deep learning has made progress and has high recognition accuracy.Compared with the computer vision algorithm,the object recognition system is closer to hardware and users.Based on the existing image recognition technology,this paper focus on the techniques needed for object recognition using the images collected by IoT devices.This paper focuses on computing distribution,data transmission and incremental learning.The main work of this paper is as follows:(1)This paper proposes an incremental image recognition framework suitable for the IoT environment.The framework separates the processes into feature extraction,training and inference.These three parts can be allocated to different IoT devices as independent modules.Reasonable allocation of these three modules allow the system to incremental learning and improve system performance.(2)This paper proposes an algorithm to allocate the operation of neural network inferencing process into multi-layer IoT devices to accelerate object cognition.Different from traditional object cognition technology,image recognition algorithm needs more resources,which makes it too expensive to complete all the processing in the data acquisition device.However,IoT devices are independent,and data transmission between devices will consume resources.This algorithm can obtain the optimal allocation strategy,which can speed up the object cognitive inference process deployed in the IoT environment.(3)This paper design and implement an object cognition system on an embedded system based on the technology above.The object cognition system proposed is an open system with wide adaptability.Users can design the specific functions of the application.This highly interactive pattern can provide users with personalised solutions at a low cost.