Neural Network Training With Brightness Correction for Hyperspectral Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hyperspectral images captured under varying lighting conditions lead to changes in spectrum, causing false detection and over-detection issues in neural network models due to inconsistent brightness.
Innovation Solution
A method and device for training a neural network model using a correction constant calculated through algorithms like GrayWorld or Gray-Edge, converting input data to RGB, detecting edges, and applying the correction constant to stabilize brightness, ensuring consistent input quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If hyperspectral images are captured under varying lighting conditions, then the imaging process can be performed in different environments, but the spectrum changes causing false detection and over-detection issues
Solution Approach 1:
The patent applies preliminary action by calculating correction constants before neural network inference. The system pre-processes calibration data to generate correction constants that compensate for lighting variations, so when hyperspectral images are captured under varying lighting conditions, the pre-calculated correction constants are already available to correct the spectrum and maintain detection accuracy without requiring real-time adjustments during imaging
Solution Approach 2:
The patent applies parameter changes by modifying the spectral parameters of hyperspectral images using correction constants. The system changes the spectral parameters through mathematical correction based on calibration data, transforming the affected spectral parameters into corrected values that maintain consistent detection performance across different lighting conditions while preserving the adaptability to image in different environments
2Measurement precision
If correction constants are calculated using Gray-Edge algorithm with edge area detection, then brightness correction accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the image processing into distinct stages: edge detection, edge area extraction, and correction constant calculation. The Gray-Edge algorithm segments the processing workflow to focus computational resources on edge regions that contain the most informative data for brightness correction, thereby improving accuracy while managing complexity through structured division of the processing task
Solution Approach 2:
The patent uses the correction constant as an intermediary element that mediates between the raw hyperspectral data and the final detection results. The correction constant serves as a pre-calculated intermediary that encapsulates the complex edge-based analysis, allowing the actual correction process to be a simple parameter application rather than repeated complex computations, thus improving accuracy while reducing real-time processing complexity
Data Source
AI summary
Disclosed is a method of training a neural network model, performed by a neural network model training device, the method including converting input data to a red, green, blue (RGB) image; calculating a correction constant based on a generation method of the input data; and training the neural network model using the correction constant.