Image segmentation method with improved boa and scheduling processing

Abstract

Image segmentation is a key step in image processing and analysis. However, traditional segmentation methods suffer from slow convergence and low segmentation accuracy. Deep learning methods, such as mask region-based convolutional neural networks, can segment medical images with an accuracy of more than 90% but require high-performance computing hardware.To solve this challenge, this study uses chaotic sequences and scheduling to change the population’s initial position, and uses adaptive weights and dynamic thresholds to adjust the individual search method. Moreover, a memory function is introduced to record the optimal position, and elite reverse learning is used to improve the population’s quality.The convergence of the enhanced algorithm has improved by an average of 46.1%, with significantly better convergence accuracy than traditional algorithms. Compared with the butterfly optimization algorithm, the improved algorithm had an average lower RMSE and Hausdorff distance by 0.58 and 0.06, and an average higher IoUR and set similarity by 0.148 and 0.248, respectively.The results show that the improved model can enhance the convergence, optimal optimization accuracy, feature similarity, and IoUR, and reduce the root mean square error of segmented images. The improved algorithm provides a reference for subsequent image processing research.