Research And Implementation Of Small Objection Flaw Surface Detection System Based On Deep Learning

With the development of wireless sensor equipment and intelligent terminal equipment technology in recent years,some inherent problems in traditional industrial manufacturing are gradually disintegrating.By relying on information control technology,combining current Internet emerging technologies,combining Internet technology with traditional industrial control technology,it can provide more fine-grained and more flexible operations for the original equipment automation control,factory production equipment status monitoring,etc.,and realize production Improved resource utilization.At the same time,the development of deep learning technology has also provided more complete technologies for intelligent manufacturing and defect detection.However,there are still many problems with intelligent defect detection systems:(1)Traditional convolutional neural networks are commonly used as feature extraction networks.While improving the network fitting ability,the pixel information of the small object is lost,and the accuracy of the small object detection is insufficient.(2)The use of edge computing for defect detection requires a huge amount of calculations and requires very high hardware,resulting in the current cost of intelligent defect detection systems.In order to solve the above problems,this paper proposes a small object surface defect detection system based on deep learning technology.This paper proposes an algorithm to improve the detection performance of small objects through multi-dilated rate convolution and fusion of multiple receptive field feature maps,and realize the function of automatic detection of product defects in the system.The algorithm changes the convolutional layer inherent in the traditional object detection network,and obtains the feature map of multiple receptive fields by combining multiple dilated convolutions with different dilated rates,which alleviates the traditional object detection network's lack of resolution and perception of small objects.The problem of lack of field has been proved by simulation experiments to have a better effect on small object detection.Our main work is:(1)through the investigation of the actual requirements of product defect detection,formulate the design and structure of the intelligent detection system based on deep learning;(2)through the investigation of the current mainstream small object detection algorithm,propose a method of network structure and training scheme to enhance the detection accuracy of small objects by dilated convolution,and to verify the detection effect on small object datasets through simulation experiments;(3)Taking the small object detection algorithm as the core,we build a product defect detection system based on the edge cloud collaboration.Finally,the small object surface defect detection system proposed in this paper can achieve dual control of the product production line in the cloud remote and on-site,and improve the detection accuracy of small target defects through a small object detection network based on multiple dilated convolution.It also realizes the functions of product surface image acquisition,product surface defect detection,product defect alarm,and product quality detection data analysis,which greatly reduces the operating costs of management personnel.