Chemical Structure Image Recognition Across Drawing Styles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis techniques struggle to accurately identify structural formulas of compounds from images due to variations in drawing formats, making it difficult to search for such data effectively.
Innovation Solution
An image analysis apparatus and method using machine learning to generate symbol information from structural formula images, employing a convolutional neural network for feature extraction and a recurrent neural network for symbol recognition, capable of adapting to different drawing styles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If rule-based identification methods are used to recognize structural formulas, then identification accuracy is maintained for standard formats, but the system cannot adapt to new or varied drawing styles
Solution Approach 1:
The patent transforms the identification approach from rule-based to machine learning-based, where the system learns to recognize structural formulas by adjusting internal parameters (weights and biases in neural networks) rather than following explicit rules. This allows the system to adapt to various drawing styles without requiring additional rules for each style.
Solution Approach 2:
The patent replaces the mechanical rule-based identification system with a machine learning system that automatically learns patterns from training data. Instead of manually defining rules for each drawing convention, the system uses neural networks to automatically identify structural formulas in various formats.
2Adaptability or versatility
If multiple rules are established to cope with different drawing formats, then coverage of various formats is improved, but system complexity and maintenance difficulty increase
Solution Approach 1:
The patent applies preliminary action by training the machine learning model on diverse training data that includes various drawing styles and formats before actual use. This pre-training process enables the system to handle multiple formats without requiring separate rules for each format, simplifying both development and maintenance.
Solution Approach 2:
The patent creates a universal identification system using machine learning that can handle multiple drawing formats with a single model. Instead of creating separate rule sets for different formats, the neural network learns to recognize structural formulas across various styles, making the system multi-functional and easier to maintain.
3Measurement precision
If traditional pattern recognition is used for text information extraction, then processing speed is maintained, but accuracy decreases for varied structural formula representations
Solution Approach 1:
The patent replaces traditional pattern recognition methods with machine learning-based image recognition. The system uses neural networks to automatically learn features and patterns from training images, achieving higher accuracy in identifying structural formulas while the system manages the complexity of the analysis model through automated training processes.
Data Source
AI summary
There are provided an image analysis apparatus, an image analysis method, and a program for implementing an image analysis method that can, when text information about a structural formula of a compound is generated from an image showing the structural formula, cope with a change in the way of drawing of the structural formula. An image analysis apparatus according to one embodiment of the present invention includes a processor, and the processor is configured to generate, on the basis of a feature value of a subject image showing a structural formula of a subject compound, symbol information representing the structural formula of the subject compound with a line notation, by using an analysis model. The analysis model is a model created through machine learning using a learning image and symbol information representing a structural formula of a compound shown by the learning image with a line notation.