Text Recognition of Raster Geological Images Based on Improved EAST and CRNN Models
A large number of raster geological images contain rich geographical and geological information.Efficient extraction of text from these images aids in data mining,as well as image retrieval and management.This paper proposes a text localization and recognition method based on improved EAST and CRNN models.Firstly,a quality focal loss method is introduced in the EAST model to enhance the focus on sample difficulty,classification errors,and the consistency of regression errors,and image segmenta-tion is performed according to the scale.Secondly,the CRNN model adopts the more lightweight and efficient MobileNetv3 as the backbone network,combined with a feature map fusion technique of different sizes.Experimental results show that the proposed im-provements effectively enhance the model's performance,achieving F1-Scores of 0.860 and 0.808 in text localization and recogni-tion tasks for raster geological images,respectively.
NETL
NSTL
万方数据
