Fiber Image Segmentation Correction for Overlapping Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation technologies struggle to accurately identify individual fibers, especially when they are entangled or overlapping, and often misidentify line-shaped dirt as fibers.
Innovation Solution
An image processing system utilizing a combination of instance and semantic segmentation models, specifically Mask R-CNN and DeepLab, to generate and correct individual-object segmentation results by merging them with category segmentation results through logical conjunctions and adjustments like dilation or smoothing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If individual-object segmentation model is used to detect each fiber, then fiber detection speed is improved, but accuracy deteriorates when fibers are entangled or overlapping
Solution Approach 1:
The patent combines individual-object segmentation model and category segmentation model into a hybrid system. The individual-object model provides fast fiber detection, while the category segmentation model provides accurate region identification. By merging their results through logical conjunction and score-based selection, the system achieves both speed and accuracy, resolving the contradiction between detection speed and identification accuracy.
Solution Approach 2:
The region correction part acts as an intermediary that processes and refines the segmentation results. It calculates logical conjunction of individual-object and category segmentation results, compares scores, and generates corrected individual-object segmentation results. This intermediary processing step enables the system to maintain fast detection while improving accuracy for entangled and overlapping fibers.
2Productivity
If individual-object segmentation model is used to detect fibers, then detection efficiency is improved, but false detection of line-shaped dirt increases
Solution Approach 1:
The system merges individual-object segmentation results with category segmentation results through logical conjunction. The category segmentation model provides contextual information about fiber regions, helping to distinguish real fibers from false detections like line-shaped dirt. This combination maintains detection efficiency while improving reliability by filtering out false positives.
Solution Approach 2:
The region correction part uses score comparison as feedback to refine detection results. By comparing scores from individual-object and category segmentation models, the system can identify and correct false detections. The feedback mechanism allows the system to maintain high detection efficiency while reducing false detection rate through iterative refinement.
3Measurement precision
If category segmentation model is used to recognize fiber regions, then accuracy for overlapping fibers is improved, but individual fiber identification capability deteriorates
Solution Approach 1:
The patent applies segmentation by using two different segmentation models with complementary strengths. The individual-object segmentation model segments each fiber individually for detection, while the category segmentation model segments fiber regions for accurate recognition. By segmenting the processing into these two specialized models and combining their results, the system achieves both individual fiber detection capability and accurate region recognition.
Solution Approach 2:
The system merges the outputs of individual-object and category segmentation models through logical conjunction and score-based selection. This merging process allows the individual-object model to provide fast individual fiber detection while the category segmentation model ensures accurate region recognition, resolving the contradiction between detection capability and recognition accuracy.
Data Source
AI summary
An image processing device includes an image acquisition part configured to acquire a target image in which fibers are captured, a first segmentation part configured to generate an individual-object segmentation result detecting each of the fibers included in the target image using a trained individual-object segmentation mode, a second segmentation part configured to generate a category segmentation result recognizing regions where the fibers are captured in the target image using a trained category segmentation model, a region correction part configured to correct the individual-object segmentation result with the category segmentation result, and a result output part configured to output a correction result of the individual-object segmentation result.