Spreadsheet Table Detection Using Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual identification of tables in spreadsheets is time-consuming and resource-intensive, especially when dealing with complex formatting, merged cells, and varying orientations, making it challenging to accurately detect and classify table locations and cell types.
Innovation Solution
A computer-implemented method using machine learning models, specifically a trained table detection model and cell classification model, to automatically identify table locations and cell types by extracting features from spreadsheet cells, performing contour detection, and classifying cells as table or background cells, and determining cell structure types such as header, data, or derived cells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual identification methods are used to detect tables in spreadsheets, then users can identify table locations and cell types, but the process becomes time-consuming and resource-intensive
Solution Approach 1:
The patent replaces manual mechanical identification processes with automated machine learning models. The table detection model and cell classification model automatically analyze spreadsheet images, extracting features and predicting table locations and cell types without human intervention, thereby eliminating time loss while maintaining or improving detection accuracy
Solution Approach 2:
The system enables spreadsheets to self-identify their table structures through automated machine learning analysis. The models process the spreadsheet image, detect table boundaries, and classify cells autonomously, allowing the system to serve itself without requiring manual user input or intervention for table detection
2Productivity
If automated machine learning models are used to detect tables, then time and resources are reduced, but the system complexity increases
Solution Approach 1:
The patent segments the table detection task into distinct specialized models: a table detection model for identifying table locations and boundaries, and a cell classification model for categorizing cell types. This segmentation allows each model to focus on specific functions, improving overall productivity while managing complexity through modular architecture
Solution Approach 2:
The machine learning models are designed to perform multiple functions: detecting table boundaries, identifying cell locations, and classifying cell types (header, data, total, grand total). This multi-functionality increases productivity by handling various table structures with a unified system rather than requiring separate solutions for each function
3Adaptability or versatility
If complex formatting, merged cells, and varying orientations are handled, then more table types can be detected, but the detection difficulty increases
Solution Approach 1:
The patent employs feature extraction techniques that transform various table formatting parameters (merged cells, orientations, formatting styles) into standardized numerical features that the machine learning models can process. By converting diverse visual parameters into a common feature space, the system achieves high adaptability to different table formats while managing detection difficulty through consistent feature representation
Data Source
AI summary
The present disclosure involves systems, software, and computer implemented methods for automatically identifying table locations and table cell types of located tables. One example method includes receiving a request to detect tables. Features are extracted from an input spreadsheet and provided to a trained table detection model trained to predict whether worksheet cells are table cells or background cells and to a cell classification model that is trained to classify worksheet cells by cell structure type. The table detection model generates binary classifications that indicate whether cells are table cells or background cells. A contour detection process is performed on the binary classifications to generate table location information that describes at least one table boundary in the spreadsheet. The trained cell classification model generates a cell structure type classification for each cell that is included in a table boundary generated by the contour detection process.