Research on Handwritten Marking Localization Method for Hot Rolled Steel Coils

Aiming at the problem that the handwritten markings of steel coils in the natural scenario of hot rolled steel coils warehouse in the travelling link using the existing text localization method have serious distortion and blurring, thus leading to a high leakage rate, a precise localization method for the handwritten markings of hot rolled steel coils is proposed. The method is based on Nanodet-m improved Coord-Nano steel coil handwritten labeling localization model. First, the backbone network perceives the spatial information of the feature map by introducing CoordConv and SAConc, which improves the adaptability and accuracy of the network in dealing with features of different scales and complexities. Then, the PfAAMLayer attention mechanism is introduced to produce a self-focusing effect on the features, which improves the architectural performance. Finally, the bottom-up path enhancement module is utilized to propagate the strong localization features in the shallow network to the deep network, so as to accurately locate the handwritten marking region of the steel roll from the complex environment. Experimental results show that the network model proposed in this paper achieves 97.7% localization accuracy for steel roll handwritten markers by training and validation on real steel roll handwritten markers image dataset.Compared with the basic Nanodet-m, YOLOv7tiny and YOLOv8n, the localization accuracy of handwritten marks on steel rolls is improved by 1.6%, 0.8% and 0.2%.