Research On Surface Defect Detection Technology Of Aluminum Profiles Based On Deep Learning

Aluminum profile is the most commonly used profile in our daily life,which is widely used in the fields of aviation,construction,industry,home decoration doors and windows,etc.Due to its huge demand,the production and sales volume of aluminum profile in China has become an increasing trend year by year.In different application fields,the quality requirements for aluminum profiles are different.In the production process of aluminum profiles,due to various factors such as mechanical friction,product technology,raw materials and other aspects,it will lead to different types and different degrees of defects in the produced aluminum profiles.At present,aluminum profile manufacturers generally adopt manual inspection for the detection of defects in aluminum profile products,but this inspection method has large limitations and is highly subjective,which cannot form a unified judgment guideline and cannot adapt to the requirements of modern industrial production.In recent years,with the rapid development of machine vision and deep learning technology,the surface defect detection method based on machine learning and deep learning provides a new way of thinking for automated quality inspection of aluminum products.Therefore,this thesis combines machine vision with deep learning and proposes a deep learning-based surface defect detection system for aluminum profiles.Firstly,a surface defect detection system for aluminum profiles is designed according to the requirements of the inspection task.The system mainly includes four parts: light source module,image acquisition module,transmission module and sorting module,which realize the acquisition of image information on the surface of aluminum profiles and the sorting of normal and defective products.Secondly,on the basis of the existing public data,the data set required for the aluminum profile surface defect detection model is constructed.The analysis of the types and shapes of defects on the surface of aluminum profiles shows that the categories of surface defects mainly include 10 categories such as scuffing,bruising and bottom leakage,etc.The defects are characterized by large scale differences and irregular shapes.According to the characteristics of the defects on the surface of aluminum profiles,this thesis proposes a Faster-RCNN-based algorithm for classifying surface defects of aluminum profiles,and in order to improve the feature extraction ability of the model,a frequency-weighted noise filtering pyramid is proposed on the basis of the feature pyramid,and the high and low frequency feature weighting and noise suppression methods are used to improve the characterization ability of the model extracted to the features;the deformation convolution is used to improve the extraction ability of the model for irregular features;the extraction ability of the model for irregular features is improved by the use of the deformation convolution.A multi-stage training method is proposed to improve the classification accuracy of the model for defects by making full use of the confrontation samples generated by the model itself,and the average classification accuracy of the improved model can reach 97.93% through the test set verification,which has better defect classification effect.In this thesis,we propose an improved cascade structure model for detecting defective areas on the surface of aluminum profiles,using a network model with improved cascade structure on the basis of the original Faster-RCNN and introducing difficult sample mining to improve the detection effect of the model for defective areas of different sizes.A random multiscale scaling input strategy is proposed to improve the detection effect of the network for multiple defects in the same area,and the overall m AP of the model reaches 89.29%after experimental validation.Finally,the QT software is used to write the terminal host interface of the aluminum surface defect detection system,which integrates the detection device and the detection algorithm;the commands issued by the host can control the transmission of the detection device,and the camera of the detection device can be called by Open CV.The system terminal realizes the integrated operation of image acquisition,image processing,image detection and result storage,which reduces the difficulty of system operation.After experimental verification,the research in this thesis has achieved the expected requirements and improved the automation of aluminum profile surface defect detection to meet the actual industrial production deployment and operation requirements.