<![CDATA[The process of underwater image segmentation is also very difficult because the data collected by the underwater sensors and cameras is of very high complexity, and much data is generated and in that case, the data is not well seen, the color is distorted, and the features overlap. Current solutions, including K-means clustering and Random Forest classification, are unable to partition complex underwater images with high accuracy, or are unable to scale to large datasets, although the possibility of dynamically optimizing the number of clusters has not been fully explored. To fill these gaps, this paper advises a hybrid solution that combines K-means clustering, Random Forest classification and the Simulated Annealing optimization as a complete end to end system to maximize the efficiency and accuracy of segmentation. K-means clustering first divides images based on pixel intensity, Random Forest narrows its segmentation of images with features like texture, color and shape, and Simulated Annealing determines the desired number of clusters dynamically to segment images with minimal segmentation error. The segmentation error of the proposed method was 30 less than the baseline K-means segmentation accuracy of 65 percent and the proposed method segmentation accuracy was 95% with an optimal cluster number of 10 and a mean error of 7839.22. This hybrid system offers a large-scale, scalable system to underwater image processing that is robust and has applications in marine biology, environmental research, and autonomous underwater system exploration.]]>
Copyrights © 2025
