Research On Surface Defect Detection Algorithm Of Aluminum Profile Based On Machine Learning

With the rapid development of Chinese automation technology,Chinese manufacturing industriesface a transition from traditional manufacturing to intelligent manufacturing.Product defect detection is an important procedure for enterprises to ensure product quality.With the continuous development of computer technology,especially the improvement and application of computer vision and deep learning algorithm theory,more and more companies are applying computer vision to product defect detection,improving equipment automation,reducing employment costs,and increasing accuracy and efficiency of product defects detection.This paper studies the defects on the surface of aluminum profiles and designs a defect detection algorithm for the characteristics of aluminum surface defects.The main research contents are as follows:Firstly,the aluminum profile defect detection algorithm based on traditional image processing method and machine learning algorithm SVM was studied.The traditional image processing methods are introduced in detail,including image filtering preprocessing,image threshold segmentation,image defect feature extraction,etc.,and the principle of machine learning algorithm SVM is introduced.Then we focus on the original defect detection Gaussian-yolov3 algorithm to make targeted algorithm structure modifications.First,based on the original Gaussian-yolov3 model,in order to enhance the model's accuracy rate of small defects on the aluminum surface,we used dense connection technology to enhance featuresusage and propagation efficiency.Secondly,in order to adapt to the defects of various shapes on the surface of aluminum profiles,in order to improve the feature extraction ability of convolution,a deformable convolution technology with double adjustment was proposed;thirdly,to improve the the adaptability of the convolution kernel to image flipping,a skeletal convolution technique was proposed,which is fused with the traditional convolution.Then,in the training model phase,in order to make the model faster and more stable to converge to the global optimal solution,Radam gradient descent method was used instead of the adam gradient descent method,and a dual-cycle gradient descent test algorithm is used to further improve the defect detection accuracy.In order to solve the problem of insufficient number of defects in the original aluminum profile,Bayesian model was used to automatically find the best data enhancement strategy.Finally,this article compares the effectiveness of the improved algorithm with common target detection algorithms.The detection accuracy of improved algorithm is better than common target detection algorithms,and the speed reaches the real-time requirements.By setting up improvement scheme comparison experiments,the effectiveness of the improvement scheme is verified.By comparing the improved algorithm with the traditional image processing algorithm SVM,the accuracy of the improved algorithm is better than the SVM algorithm.The experiments verify that the algorithm designed for the surface defects of aluminum profiles in this paper can meet the requirements of industrial inspection accuracy and has high practical value.