Automatic Condom Loading Device Design Based On Seminal Vesicle Image Recognition

The demands for disease protection and changing conceptions of reproduction have led to a yearly worldwide condom usage of roughly billions.The quality inspection procedure for condoms currently requires workers to frequently load and operate an electric dry inspection machine.This wears down the workers’ bodies and negatively impacts their mental health due to the monotonous working environment.This thesis develops a set of automated condom examination and feeding devices to tackle this issue,leaving the device to handle the tedious task on its own.Firstly,an automated loading device based on condom picture sperm vesicle recognition was developed by the electrical dry examination machine’s operating principle.The gadget primarily consisted of an industrial camera,a programmable logic controller(PLC),a condom-grasping mechanism,a condom-taking mechanism,a condom-feeding mechanism,and a stage for displaying condoms.The condom pick-up mechanism removed the condom for testing from the container or tray.The condom was then transported by conveyor belt to the image acquisition area.The industrial camera took a picture of the condom and send the data to the main control system for identification.The main control system sent the identified condom sperm sack position information to the sub-control system;the sub-control system controls the robot arm to grab the condom and transported it to the feeding mechanism with the opening facing down;the feeding mechanism places the condom into the mold bar.Secondly,the condom seminal vesicle region was located using the conventional target recognition algorithm.The original image must be properly preprocessed for contour extraction before the detection of the seminal vesicle area because the contour shape of the condom was distinctive.In this thesis,the improved Canny edge detection algorithm was used to obtain the edge map of the condom image with higher edge detection accuracy.However,industrial cameras used to take pictures would have an impact on the subsequent picture detection due to noise interference caused by conveyor belt transport condoms,such as silica.For instance,changes in the external environment could make it difficult to choose a threshold range,which meant that after testing and verification,the conventional target detection algorithm was not well suited to the site process requirements.Thirdly,to tackle aforementioned issue,Yolov5 algorithm was proposed to train the images.The sperm vesicle area of the acquired images was subjected to feature extraction in order to produce image features,and the data sets were enhanced by a better version of mosaic-9 data enhancement for increasing the number of tiny target samples and accelerating network training.To avoid overfitting the model,the Label-Smoothing algorithm was put forward.Additionally,a SENet attention method was added to enable the network to give more attention to crucial feature channels,and enhance network functionality to produce weighted feature maps.Lastly,the upper computer system was then constructed and empirically validated.The ONNX inference machine was ultimately chosen as the training model for this thesis through analysis of different model frameworks,and the identified pictures and feature coordinates were presented to the Qt upper computer.The experiment results show that the recognition rate and detection speed of the condom detection device satisfy the field detection criteria.