Research On Image Recognition Technology In Internet Of Things System

With the development of the Internet of Things technology,there are more and more application scenarios for collecting image data using cameras.However,in order to improve the efficiency of data transmission,most of the images are compressed and processed,with low definition and a certain amount of noise.Here comes a great challenge.At the same time,with the development of computer technology and image recognition technology,artificial intelligence technology is gradually applied to people's production and life.How to use artificial intelligence and image recognition technology to achieve efficient and accurate recognition has become a hot issue in current research.This paper conducts research on the image recognition technology in the Internet of Things system,studies the depth-based image recognition technology,constructs the meter reading and multi-commodity recognition image data set,builds a deep network model,and realizes the detection and recognition of meter reading Commodity detection and identification,the main work and results of the paper are as follows:1.Researched the image recognition technology in the Internet of Things system,and researched and analyzed the currently commonly used target detection algorithms,including traditional detection and recognition methods,mainly related to digital image processing technology,and detection based on artificial intelligence Recognition methods,including SVM,neural network and deep learning technologies.2.Using multi-layer convolutional neural network and transfer learning technology to achieve accurate detection and recognition of meter readings.Acquire and label the meter reading image data set through the camera,and expand the data set through image processing and enhancement techniques,build a multi-layer convolutional neural network,and combine transfer learning and other technologies to achieve the accuracy of meter reading detection and recognition.98.0%.3.The improved SSD algorithm is adopted to realize the multi-target detection and recognition of commodity images.Obtain and annotate the data set through the camera,and combine the image processing and enhancement technology to expand the data set,improve and optimize on the basis of the SSD algorithm,and use the expanded convolution to replace multiple convolutional layers for the problem of excessive model size To reduce the parameters by reducing the number of layers;for the problem of poor detection of small targets in the image,add a feature pyramid structure(FPN)to the model structure to improve the detection effect of small targets.After the above optimization,mAP reached 80.0 and FPS reached 36.