Research On Recognition Technology Of Micro-Defects On The Surface Of Aluminum Profiles For Unbalanced Samples

Industry is the foundation of the country’s economy.Aluminum profile is an important product of industrial profiles.Workpiece defect detection is an important process in its production process.Surface fine defect detection is an important work for workpiece defect detection,which affects the practicality and appearance of aluminum profile.Sex and comfort.Due to the poor characterization of fine defect images,it is difficult to collect samples,which leads to the imbalance of sample space and the difficulty of identifying fine defects.However,there is currently no effective method to identify the surface fine defects in the unbalanced sample space.In order to improve the recognition accuracy of the micro-defects in the unbalanced sample space,this paper takes the surface defects of aluminum profiles as the research object,and proposes a method of identifying the micro-defects on the surface of aluminum profiles for the unbalanced sample space,and builds a data equalization model and super-resolution Fusion classification model,the specific research content and main conclusions are as follows.In order to improve the identification accuracy of the micro-defects in the uneven sample space,this paper takes the surface defects of aluminum profiles as the research object,analyzes the sample data of the surface defects of aluminum profiles,and characterizes the sample space and micro-defects according to the production process of aluminum profiles.The summary is completed,the impact of imbalanced samples and subtle features on the classification accuracy is obtained,and the necessity of studying the imbalanced sample space and subtle image features is proposed.A method for identifying micro-defects on the surface of aluminum profiles for unbalanced sample spaces is proposed,and a data equalization model and a super-resolution fusion classification model are constructed.The specific research content and main conclusions are as follows:(1)Aiming at the problem of unbalanced sample space of aluminum profile industrial images,a data balancing method based on full sample background space is proposed.This method includes sample restoration,feature judgment and extraction,and equalization sampling to achieve the balance of the original sample space of aluminum profile data Change;(2)Aiming at the problem that it is difficult to identify surface minor defects,the super-resolution generation of minor defects is completed based on the generative confrontation network,including the extraction of minor defects,super-resolution restoration,feature fusion,and the construction of an attention classification model based on the residual mechanism.A defect detection model with higher classification accuracy;(3)On the basis of the above research,configure the experimental software and hardware environment platform,firstly build a classification network to verify the equalization method of the unbalanced sample space,and then realize the division of fine samples and super-resolution feature extraction on the original sample space data,and The attention mechanism model is built to verify the superiority of the detection method for subtle defects,and proves that for classification detection tasks,this method can synergistically improve the detection effect from both the sample space and the model structure.