Chemical Structure Recognition from Document Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to accurately recognize chemical information from document images, particularly chemical structures in standardized and non-standardized formats, as well as chemical reactions, due to limitations in handling noisy or poorly detailed images.
Innovation Solution
A computer-implementable method using a system comprising a detector, structure recognition unit, arrow recognition unit, and reaction recognition unit, employing neural networks such as Faster R-CNN and modified Transformer architectures to identify and classify chemical structures and reactions from document images, converting them into SMILES notation for accurate data extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing recognition methods are used for chemical structures in non-standardized formats, then processing speed is maintained, but recognition accuracy significantly deteriorates
Solution Approach 1:
The patent applies parameter changes by transforming chemical structure images into multiple augmented versions with varied parameters including different rotations (0°, 90°, 180°, 270°), flips (horizontal, vertical, both), and scaling factors (0.5, 0.75, 1.0, 1.25, 1.5). This creates a diverse training dataset that enables the neural network to accurately recognize chemical structures across various non-standardized formats and orientations, directly resolving the contradiction between recognition accuracy and adaptability to different formats
2Measurement precision
If complex neural network models are employed to improve recognition accuracy, then measurement precision improves, but computational complexity and processing time increase
Solution Approach 1:
The patent implements preliminary action through extensive data augmentation before training, where chemical structure images are pre-processed into multiple transformed versions (rotations, flips, scaling) and stored as a prepared dataset. This preliminary preparation enables the neural network to learn from diverse examples during training, achieving high recognition accuracy without requiring excessively complex model architectures or prolonged processing times during actual recognition tasks
3Measurement precision
If multiple transformation versions of chemical structures are generated for training, then recognition accuracy for varied formats improves, but data processing complexity increases
Solution Approach 1:
The patent applies self-service by implementing an automated data augmentation system that programmatically generates transformed versions of chemical structure images using defined transformation rules (rotations, flips, scaling). The system automatically creates the augmented dataset and prepares it for training without manual intervention, reducing processing complexity despite generating multiple transformation versions. The automated pipeline handles the complexity of creating diverse training data while maintaining recognition accuracy
Data Source
AI summary
The present disclosure is related to the field of data recognition. The computer-implementable method includes the following steps: inputting an image of a document page to a detector; the detector identifies fragments on the page; obtaining coordinates of the fragment on the page for each identified fragment; and classifying the fragments; the structure recognition unit recognizes the chemical structure for each fragment; inputting identified fragments of the reaction arrows to an arrow recognition unit; obtaining coordinates on the page for each arrow and reaction attributes; transmitting to an input of a reaction recognition unit the coordinates on the page for each fragment of the recognized chemical structures; and based on the obtained data the reaction recognition unit determines how the arrows relate to the recognized chemical structures; as a result, based on the recognized data for the image of the document page, obtaining recognized chemical structures.