Part Inspection Method For Industrial Production Line Based On Convolutional Neural Network

Components on industrial production lines are an indispensable part of the constituent products.In particular,exporting overseas industrial products must ensure the integrity of the components needed to make products.In actual industrial production,parts are prone to leakage and misassemble,affecting the qualified rate of products,and causing serious economic losses to the company.In recent years,most parts inspections on industrial production lines have remained in the stage of manual detection and traditional shape template matching detection,and the detection efficiency is low.In manual detection,long-term work can lead to visual fatigue of the human eye,increasing the risk of misdetection and missed detection,and to a certain extent requires a lot of labor costs.At present,with the development of artificial intelligence and industrial automation,computer vision technology is increasingly used in the field of industrial inspection,and has achieved good results,injecting new detection solutions for industrial production and improving industrial production efficiency.In order to achieve automatic detection of parts and components in industrial production lines and improve detection efficiency,this paper presents a component detection method based on deep convolutional neural network.Firstly,sample images are collected to perform sample image preprocessing operations,which mainly include image annotation and image enhancement.To ensure the diversity of training samples,physical morphing of the samples is performed,mainly including image affine transformation,image rotation,image blur,and image scaling.And so on to achieve data amplification.Then by using the deep convolutional neural network training detection model,including the extraction of effective features of the target,and on different convolutional feature maps,the predicted foreground target candidate frames are generated.Considering that the existing detection framework uses a deep convolutional neural network to calculate a large amount of parameters,the detection speed is slow in a GPU-free hardware environment,and large-scale application problems cannot be solved.This paper designs a new training network structure based on Mobile Nets lightweight network,effectively reduces the amount of network computing parameters,and reduces the size of the detection model.Finally,the confidence level of the candidate frame is calculated,and the detection target category is output.At the same time,the prediction candidate frame is subjected to a regression operation,the positioning position is fine-tuned,and the final detection result of the target component is output.The experimental results show that the proposed method for component inspection on industrial production lines effectively improves the detection efficiency.At the same time,under the premise of not affecting the detection accuracy,the network training structure model is reduced,the detection speed can be effectively improved on the small GPU-free industrial control machine,and the actual industrial inspection requirements can be satisfied,and the large-scale application is on the industrial production line.