Multispectral Image Collection for ML Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine-learning models trained with visible light images can only identify defects visible in the visible light spectrum, limiting their accuracy in detecting defects that are not visible in this spectrum.
Innovation Solution
Collect and condition image data by capturing products under various wavelengths of light, including ultraviolet and infrared, and designate defect locations across images with the same viewpoint to create a comprehensive image dataset for training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image data is captured only in visible light spectrum, then the machine-learning model can be trained with simple imaging equipment, but the model can only identify defects visible under visible light, limiting detection accuracy
Solution Approach 1:
The imaging system is segmented into multiple specialized cameras, each designed to capture specific wavelength ranges (ultraviolet, visible, infrared). This segmentation allows each camera to be optimized for its specific function while collectively providing comprehensive multi-spectral coverage, resolving the contradiction between detection accuracy and system complexity
Solution Approach 2:
The imaging system is designed with multi-functionality by incorporating cameras that can detect across multiple wavelength spectra (UV, visible, IR). This universal capability enables a single system to detect various types of defects that may be visible only in specific wavelength ranges, improving detection accuracy without requiring multiple separate systems
2Measurement precision
If multiple wavelengths of light are used to capture images, then the machine-learning model can identify defects across different light spectra, but the complexity of collecting and processing image data increases
Solution Approach 1:
Multiple images captured at different wavelengths are merged and combined into a unified image dataset. The system integrates UV, visible, and IR images by aligning them spatially and combining them into composite images or structured data formats, simplifying the processing complexity while maintaining the benefits of multi-wavelength detection
Solution Approach 2:
The system creates standardized copies and representations of defect information across different wavelength images. By identifying defects in one wavelength range and copying/propagating this information to corresponding locations in other wavelength images, the system reduces processing complexity while ensuring comprehensive defect detection across all spectra
Data Source
AI summary
A method of collecting and conditioning image data of an item for use in training a machine-learning model to detect at least one defect can include capturing a first set of images by a first camera with each image of the first set of images having a first viewpoint of the item that is the same viewpoint as the other images in the first set of images and can be illuminated by various wavelengths of light including ultraviolet light, infrared light, and visible light. The method can further include examining one image of the first set of images to determine if the item contains at least one defect, identifying a first location of the at least one defect on the one image, designating the first location on the other images of the first set of images so that all images in the first set of images identify the first location.