基于改进粒子群算法的植物冠层图像分割

Multi Threshold Segmentation of Plant Canopy Image Based on Improved Particle Swarm Optimization

针对标准粒子群算法易陷入局部最优而导致图像分割效果欠佳的问题,采用一种与模拟退火算法相结合的混合粒子群算法来优化多阈值图像分割的阈值选取过程,将Otsu类间方差函数作为算法的适应度函数,并利用模拟退火算法"突跳"的特点有效避免陷入局部最优。实验结果表明:该算法可以有效地处理复杂植物冠层图像分割的问题,能够在保证运行效率的同时提高图像的分割精度。为提高植物生长状态评估的可靠性以及叶片信息的准确性提供理论基础,具有较强的工程实用性。

Aiming at the problem that the standard particle swarm algorithm is easy to fall into the local optimum which leads to poor image segmentation effect,a hybrid particle swarm optimization algorithm combined with simulated annealing algorithm is used to optimize the threshold selection process of multi-threshold image segmentation.The variance function between Otsu classes is used as the fitness function of the algorithm and simulated annealing algorithm is used to avoid jumping into local optimum.The experiment results show that the algorithm can effectively deal with the problem of complex plant canopy image segmentation,and can improve the image segmentation accuracy while guaranteeing the operation efficiency.It provides a theoretical basis for improving the reliability of plant growth condition evaluation and the accuracy of leaf information,with a strong engineering practicability