Research on hot-rolled coil surface defects detection technology based on improved YOLOv3 model
Hot-rolled coil is an important raw material for steel production,and its surface defects di-rectly affect the quality of the final steel products.Therefore,the detection technology of surface de-fects on hot-rolled coils is of significant importance for controlling product quality.In order to achieve more accurate and efficient determination of the surface quality of hot-rolled coils,reduce labor costs in Tangshan Iron and Steel's hot-rolling production line,and thus improve production efficiency,stud-ying automatic detection technology for surface defects on hot-rolled coils is imperative.This paper proposes improvements to the YOLOv3 model:the introduction of the SE-PRE improved attention module to increase the model's focus on defect areas;the addition of a detection branch to expand the receptive field,improving the model's ability to detect small target defects;and the introduction of a dropout detection head to effectively control overfitting during training.Additionally,in terms of model training,the learning rate setting,loss function,and selection strategy for the optimal model were opti-mized.Data augmentation techniques is also used to expand the data volume.Through a large number of on-site data experiments,the results show that the model in this paper has achieved good results in both detection accuracy and efficiency.This is significant for reducing on-site labor costs and impro-ving product quality.
NETL
NSTL
万方数据
