Underwater image illumination estimation via an evolving extreme learning machine by an improved salp swarm algorithm

Abstract

Underwater images have chromatic aberrations under different light sources and complex underwater scenes, which can lead to the wrong choice when using an underwater robot. To solve this problem, this paper proposes an underwater image illumination estimation model, which we call the modified salp swarm algorithm (SSA) extreme learning machine (MSSA-ELM). It uses the Harris hawks optimization algorithm to generate a high-quality SSA population, and uses a multiverse optimizer algorithm to improve the follower position that makes an individual salp carry out global and local searches with a different scope. Then, the improved SSA is used to iteratively optimize the input weights and hidden layer bias of ELM to form a stable MSSA-ELM illumination estimation model. The experimental results of our underwater image illumination estimations and predictions show that the average accuracy of the MSSA-ELM model is 0.9209. Compared to similar models, the MSSA-ELM model has the best accuracy for underwater image illumination estimation. The analysis results show that the MSSA-ELM model also has high stability and is significantly different from other models.