Research on surface defect detection method of aluminum profile based on machine vision

Aluminum profiles are indispensable basic raw materials in the industrial manufacturing process and are widely used in various fields. However, in the production process of aluminum profiles, due to the detection environment and the production of raw materials in a variety of cases, there are often surface defects, these defects will reduce the quality of aluminum profiles, and can not meet its actual needs in the industry. Therefore, it is important to detect the surface defects of aluminum profiles. Distinguished from manual inspection means and methods that can lead to misdetection and omission and other problems, machine vision technology has been widely used in defect detection in industrial production due to its non-contact, high sensitivity and high adaptability and other characteristics. However, with the progress of society and the development of science and technology, the accuracy and real-time performance of traditional machine vision detection methods in the actual detection process become difficult to meet the current industrial production needs, with the continuous development of deep learning methods, the Yolo series algorithms based on the surface defects of aluminum profiles has become a hotspot, in order to solve the problem of poor results of the original Yolov5 model in the detection of defects on the aluminum surface, this paper proposes an improved model for the detection of surface defects in aluminum profiles, with the aim of solving the problem of poor results. In order to solve the problem that the original Yolov5 model is less effective in aluminum profile surface defects detection, this paper proposes an improved Yolov5 model called Yolov5-CA.This improved model is optimized in the following aspects: 1. Anchor frame mechanism: Yolov5-CA uses LoU (Location of Union) distance instead of the Euclidean distance of Yolov5, in order to enhance the adaptability of the model to the dataset of this paper; 2. Data enhancement: The original Mosaic-4 method is replaced by Mosaic-9, which improves the training speed of Yolov5 and enhances the performance of the model; 3. Feature extraction: Yolov5-CA adds CA (Channel Attention) attention mechanism, which helps the model to better capture key features.