U-net Model with Particle Swarm Optimization for Floating Raft Information Extraction

Extracting the sea area of Marine floating raft culture is contributed to the rational use of aquaculture resources. The availability of remote sensing technology can dynamically detect aquaculture rafts. As a classical semantic segmentation model, U - net can also be beneficial to extract the synthetic aperture radar images of marine cultivation. However, the parameters of U - net are customarily set by experts based on experience and cannot be changed dynamically. In this paper, a method for flexibly designing U - net model parameters by particle swarm optimization is presented. Treat individuals with diverse parameters as a swarm and randomly initialize the encoded information, allowing each individual to apply appraisal indi-cators for fitness evaluation. Implementing particle updates and iteratively improving model parameters in interactive learning between individuals and populations. The experimental results demonstrate that the model can obtain competitive outcomes on multiple evaluation indicators after parameter optimization.