Multiclass Classification and Defect Detection of Surfaces Using Modified-YOLO

Industrial products based on glass fiber are widely used in various industries due to their excellent properties. Steel tubes known for its strength and durability are widely used in hazardous high pressure environments such as the petroleum, chemical, natural gas and shale gas industries. Deep learning greatly improves inspection efficiency in object detection and defect detection. In this work, we use a well-known You Only Look Once version 7(YOLOv7) model to achieve accurate defects detection of steel tube and glass tube images. First, the classification of the dataset is checked using a VGG16(Visua1 Geometry Group) and EfficientNet. A Convolutional Block Attention Module (CBAM) mechanism is then integrated into the YOLOv5 backbone network to improve feature expression ability. Additionally, the Generalized Intersection over Union (GIoU) loss function is used to enable the model to learn precise object localization. Experimental results show that the modified YOLOv7 is better than the other models.