Machine Learning Based On Breathable Waterproof Caps Defect Detection System

Breathable waterproof bottle cap is a kind of cap used for packaging cosmetics,medicines and other products.It can be waterproof,leak-proof,and ensure the breathability of the products inside the bottle.Breathable waterproof bottle caps usually adopt a duallayer structure design,including the inner layer of EPDM rubber ring,GORE breathable valve,and the outer layer of plastic bottle cap body.Various types of defects may occur in the manufacturing process of this type of bottle cap,including but not limited to EPDM rubber ring turning outward and abnormal assembly of GORE valve.The defects of breathable waterproof bottle caps have a serious impact on the quality and safety of the products inside the package.Currently,manual methods are widely used for testing this type of bottle cap both domestically and internationally,which wastes a lot of human and material resources and is hard to ensure accuracy and consistency.Therefore,this article applies machine learning methods to research and apply the detection of defects in breathable waterproof bottle caps.The main work of this article is as follows:(1)Design of the detection system for defects in breathable waterproof bottle caps: First,formulate the overall operating plan and workflow of the detection system and determine that the system is mainly composed of transmission device and control module,visual inspection module and rejection module.Then,select hardware equipment according to the design needs and finally build the detection system and describe the entire system’s workflow to ensure that the system can collect all information about breathable waterproof bottle caps.(2)Binary classification study of breathable waterproof bottle caps: propose an automatic classification method for breathable waterproof bottle caps based on the improved Mobilenet v1.This method consists of two parts: image preprocessing and improved classification network model.The preprocessing steps are as follows: First,use the center scaling method to unify the image pixels to the same size;second,use the grayscale algorithm of contrast stretching and gradient optimization to convert color images to grayscale;third,perform data enhancement on the images in the original dataset to expand the sample capacity.Compared with the maximum grayscale method,the image’s classification accuracy is improved by 6.56% to 89.82% through the proposed preprocessing method.The improved network is specifically implemented by adopting Swish as the activation function strategy of Mobilenetv1 to improve the overall network framework’s recognition rate.Experimental results demonstrate that the proposed method enhances the network’s training accuracy performance and achieves a 91.75% binary classification accuracy.(3)Research on precise detection method of breathable and waterproof bottle cap defects: proposed a positioning and detection framework called AF-r CNN.The framework is an improvement based on the Faster r CNN algorithm.First of all,the backbone network in Faster r CNN was modified to Resnet50 to enhance the feature extraction capability of the entire network.Secondly,a model named dual attention mechanism was proposed,which adds CBAM attention mechanism to both the backbone feature extraction network and the RPN network to improve the model’s recognition rate.Finally,the concept of transfer learning was introduced to shorten the training time of the entire framework.Experimental results show that compared to the model that only uses Resnet 50 as the main network,the m AP indicator of the AF-r CNN model is improved by 2.7%.