Abstract:

In view of the problem that intelligent optimization image segmentation algorithm is easy to fall into local optimal and low segmentation precision, this paper presents a new multi-threshold remote sensing image segmentation algorithm, which combines the improved fractional Darwin particle swarm optimization and the two-dimensional Renyi entropy multi-threshold. The algorithm uses the particle evolution information to define the evolution factor, combines the evolutionary factor and adjusts the fractional coefficients α by using the Gauss graph function to achieve accurate calculation and fast convergence.According to the local optimal probability factor, the Levy flight random disturbance is carried out to improve the ability of the algorithm to jump out of the local optimal. At the same time, the two-dimensional Renyi entropy single threshold is extended to the multi-threshold segmentation, and the improved fractional Darwin particle swarm optimization is used to apply the two-dimensional Renyi entropy multi-threshold to the remote sensing image segmentation.The simulation results show that, compared with the other two intelligent segmentation algorithms, the segmentation algorithm has obvious advantages in detail processing and segmentation accuracy, at least 7.27% increase in PRI, 6.5% decrease in VOI, and decrease in GCE by at least 10.4%.

Key words: particle swarm optimization, two-dimensional Renyi entropy, remote sensing image segmentation, levy flight, evolutionary information, multi-thresholds