Interpret The Spatial Structure Of Handwritten Organic Molecular Formulas Based On SSD

Educational technology has developed rapidly in recent years,but the automatic extraction and interpretation of handwritten organic molecular formulas has always been a problem.The common method is still manual input into the computer.Therefore,it is a meaningful work to design an interpretation method of handwritten organic molecular structure.According to the features of the chemical formula in the middle school,this thesis proposed a method to interpret the organic structure from 4 important structures.Firstly,these objects should be detected from each offline image to determine the position of each object.After comparing commonly used object detection algorithms SSD and YOLO,considering detec-tion accuracy and task scenario,SSD was choosed as the main algorithm in the detection stage.In addition,a post-processing method is designed to improve the detection accuracy.Finally,based on the geometric relationship and domain knowledge,this thesis design an algorithm of interpreting the spatial structure of handwritten organic molecular formulas.The model was evaluated in self-collected dataset consisting 2100 samples and have a good result.The detection accuracy of benzene and ring was 100%,and the accuracy of chain and double bond was more than 95%.The whole algorithm of two stage,combined with the detection stage and interpretation stage,can achieve 89.5%accuracy.