Illumination correction via support vector regression based on improved whale optimization algorithm

Variations in illumination lead to serious errors in the evaluation of chromatic aberration. We propose an illumination correction model based on the opposition‐based learning improved whale optimization algorithm for support vector regression optimization, named "OBL‐IWOA‐SVR." Because th...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Color research and application Ročník 46; číslo 2; s. 303 - 318
Hlavní autoři: Wang, Chao, Zhu, Zefei, Chen, Sheng, Yang, Junyi
Médium: Journal Article
Jazyk:angličtina
Vydáno: Hoboken, USA John Wiley & Sons, Inc 01.04.2021
Wiley Subscription Services, Inc
Témata:
Accuracy
Adaptive algorithms
Convergence
Error analysis
Illumination
illumination correction
Machine learning
Mathematical models
opposition‐based learning
Optimization
Optimization algorithms
Parameters
Population
Regression analysis
Support vector machines
support vector regression
whale optimization algorithm
ISSN:0361-2317, 1520-6378
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Variations in illumination lead to serious errors in the evaluation of chromatic aberration. We propose an illumination correction model based on the opposition‐based learning improved whale optimization algorithm for support vector regression optimization, named "OBL‐IWOA‐SVR." Because the initial population quality of the whale optimization algorithm has a significant impact on the solution speed and accuracy, the Opposition‐based learning strategy is adopted in this article to mix the original population and its opposite individuals and select the best as the new population, replacing the random initialization to generate a more suitable initial population. This increases the diversity of the population and thus overcomes the impact of the quality of the initial population on the algorithm performance. Secondly, the proposed algorithm adopts the adaptive weight and the convergence factor based on the variation of the cosine law to balance the algorithm's global exploration ability and local development ability and enhance the convergence accuracy. Finally, the algorithm utilizes good global searching ability to optimize the penalty factor and nuclear parameters and obtains the optimal support vector machine parameter combination to construct the illumination correction model OBL‐IWOA‐SVR accurately and robustly. Experimental results show that the illumination correction model proposed in this article is found superior to other models in significance analysis: the root mean square error of the proposed model is 0.0173, the smallest of all illumination correction models. Furthermore, the model exhibits good stability and high illumination estimation accuracy.
Bibliografie:Funding information
Fundamental Research Funds, Grant/Award Number: GK199900299012‐026
ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0361-2317
1520-6378
DOI:10.1002/col.22601