| 摘要: |
| 【目的】对林木冠层图像采用NCSPSO－AFSA优化支持向量机(SVM)进行图像分割，提取树干分割图，以进一步提高分割效果。【方法】对现有的小生境和交叉算子的粒子群算法(NCSPSO)进行优化，并与人工鱼群算(AFSA)混合，寻找最优惩罚系数C和高斯核函数中的参数γ；然后运用SVM方法对训练样本进行综合训练，以建立最佳分类模型；最后对香樟树、马褂木和杨树的冠层图像进行分割，并与AFSA算法、NCSPSO算法的分割效果进行比较。【结果】AFSA、NCSPSO、NCSPSO-AFSA算法的平均运行时间分别为178.909，154.661和97.213 s，平均分割准确率分别为90.83%，94.08%和98.90%，表明改进的NCSPSO-AFSA混合算法在效率上较其他2种算法提高了63%以上，而且分割准确率提高了5%～8%。【结论】运用NCSPSO－AFSA优化SVM方法对林木冠层图像进行树干图像分割，可得到最佳分割效果。 |
| 关键词: 林木图像分割 NCSPSO 人工鱼群 支持向量机 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金项目（31300472）；江苏省自然科学基金项目（BK2012418） |
|
| Segmentation of forest canopy image based on NCSPSO－AFSA optimized SVM |
|
ZHANG Dong,LIU Jun-yan,XUE Lian-feng,et al
|
| Abstract: |
| 【Objective】NCSPSO-AFSA optimized support vector machine (SVM) algorithm was adopted for image segmentation and extraction of trunk segmentation diagram.【Method】Niche and Crossover Selection Particle Swarm Optimization (NCSPSO) was optimized before being mixed with Artificial Fish Swarm Algorithm (AFSA) to seek for the best penalty coefficient C and parameter of Gauss kernel function.Then,SVM was used to establish the optimum disaggregated model of training examples.Forest canopy image segmentation test was conducted for Camphor tree,Liriodendron Chinese and Poplar and the results were compared with AFSA and NCSPSO.【Result】The average running times of AFSA,NCSPSO and NCSPSO-AFSA were 178.909,154.661 and 97.213 s,and the average accuracy rates of segmentation were 90.83%,94.08%,and 98.90%,respectively.The efficiency and accuracy rate of the optimized NCSPSO-AFSA mixed algorithm were enhanced by 63% and 5%-8%,repsecitvely.【Conclusion】NCSPSO-AFSA optimized SVM can get best consequence for trunk image segmentation based on forest canopy image. |
| Key words: forest image segmentation NCSPSO artificial fish swarm algorithm support vector machine |
