Near-Infrared Spectral Wavelength Selection Using Improved Team Progress Algorithm
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current near-infrared spectral analysis technologies face challenges in selecting optimal wavelengths for prediction models, leading to insufficient prediction accuracy and increased complexity, particularly due to issues with existing algorithms like PCA and GA, which either eliminate effective variables or suffer from premature convergence.
Innovation Solution
An improved team progress algorithm (iTPA) is introduced, which divides spectral wavebands into elite, ordinary, and garbage collection groups based on evaluation values, allowing for iterative learning and exploration behaviors to update wavelength points, thereby selecting a reduced number of wavelengths that maximize prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If principal component analysis (PCA) is used for wavelength selection, then the model volume is reduced, but effective variable information is eliminated resulting in insufficient prediction accuracy
Solution Approach 1:
The TPA extracts only the most informative wavelength variables from the full spectral data, selecting a small subset of key wavelengths that contain the most predictive information. This extraction approach reduces model complexity while preserving the essential information needed for accurate predictions, avoiding the elimination of effective variables that occurs with PCA.
2Measurement precision
If genetic algorithm (GA) is used for wavelength selection, then optimization results are improved, but the number of selected wavelengths increases and calculation complexity increases
Solution Approach 1:
The TPA segments the wavelength selection process into two distinct phases: an exploration phase that broadly searches the wavelength space, and an exploitation phase that refines the selection of informative wavelengths. This segmentation allows the algorithm to achieve good optimization results while controlling the number of selected wavelengths and reducing overall calculation complexity compared to standard GA approaches.
3Measurement precision
If genetic algorithm (GA) is used for wavelength selection, then optimization results are improved, but premature convergence occurs due to lack of group diversity
Solution Approach 1:
The TPA dynamically adjusts the balance between exploration and exploitation throughout the iterative process. In early iterations, the algorithm emphasizes exploration to maintain diversity and avoid premature convergence. As iterations progress, it gradually shifts toward exploitation to refine the wavelength selection. This dynamic adaptation ensures both reliable convergence and avoidance of premature termination.
4Device complexity
If team progress algorithm (TPA) is used for wavelength selection, then algorithm simplicity is improved, but prediction accuracy needs further improvement due to insufficient information capture
Solution Approach 1:
The improved TPA modifies key parameters of the original algorithm, including the evaluation function that assesses wavelength information quality, the update rules for wavelength selection, and the iteration control parameters. These parameter changes enable the algorithm to capture more predictive information while maintaining its simplicity and low computational complexity.
Data Source
AI summary
The disclosure discloses a method for near-infrared spectral wavelength selection based on an improved team progress algorithm (iTPA), belonging to the field of near-infrared spectral detection. The method includes: equally dividing near-infrared spectral wavebands to be selected into elite groups, ordinary groups and garbage collection groups according to evaluation values from high to low; generating a new waveband in the elite group or the ordinary group, where a wavelength point of the new waveband is selected from a random waveband in the selected group, and the wavelength point of the new waveband inherits the selected wavelength point; and enabling the inherited new waveband to select a learning behavior or an exploration behavior according to a set probability to update the wavelength point of the new waveband to generate a candidate waveband; and selecting a waveband with a highest evaluation value in the elite group as the waveband to be selected. Under the condition of ensuring the model prediction accuracy, the disclosure greatly reduces the number of wavelength variables, reduces the complexity of the algorithm at the same time, and improves the non-destructive detection accuracy in crops.