Research On The Intelligent Detection Method Of Surface Defects Of Aluminium Profiles Based On Deep Learning

Aluminium profiles are widely used in various industries as one of the important application materials in the industrial field.However,due to differences in production equipment,different production processes and other factors,different types of defects can occur on the surface of aluminium profiles during the actual production process,which can have an impact on the service life of the profiles.It is therefore essential that aluminium profiles are inspected for defects before they leave the factory.Traditional detection methods have problems such as high costs and poor generalisation capabilities,which cannot meet the needs of industrial inspection.Deep learning-based target detection algorithms can effectively solve the shortcomings of traditional detection methods.However,existing target detection algorithms have limited ability to extract defects with large differences in scale and random distribution on the surface of aluminium profiles.In addition,the method often contains a large number of parameters,which leads to a large size of the algorithm.To address the above problems,this paper aims to improve the detection accuracy and reduce the size of the algorithm,based on deep learning target detection algorithm for the detection of defects on the surface of aluminium profiles.The main research elements of the thesis are as follows:(1)To address the problem that the current mainstream target detection algorithms have low performance in detecting surface defects on aluminium profiles,this paper proposes an improved model MS-YOLOv5 based on the YOLOv5 x model.firstly,the model uses a newly designed neck PE-Neck to replace the neck of the original YOLOv5 x model in order to enhance the extraction and localisation capability of the model for defects.Secondly,a newly designed Multi-streamnet is used to replace the first detection head of the original YOLOv5 x model to enhance the model’s ability to identify defects.Experiments show that the MS-YOLOv5 model has superior detection performance compared to the current mainstream target detection algorithms.(2)In order to ensure a high detection performance with a small size and fast detection speed,this paper proposes an improved model LY-YOLOv5 based on the YOLOv5 s model.firstly,the model uses the ELU activation function instead of the activation function in the original YOLOv5 s model to improve the model’s ability to represent the feature information.Secondly,the SIo U loss function is used to replace the CIo U loss function in YOLOv5 s original model to obtain more accurate prediction edges.Finally,the Shuffle Net V2 lightweight network was used to replace the backbone network in YOLOv5 s original model and to change the depth and width of the network to reduce the size of the model.The experiments show that the LY-YOLOv5 model not only has a smaller size and faster detection speed,but also has a better detection performance.(3)In order to compare and verify the effectiveness of the lightweight LY-YOLOv5 model,an experimental platform for surface defect detection of aluminium profiles is built in this thesis.By embedding different models into the Jeston Nano development board to run,in order to test the effect of the model on the detection of surface defects of aluminium profiles.The experiments show that the LY-YOLOv5 model is optimal for the real-time detection of surface defects in aluminium profiles.