摘要
研究基于Boosting的柑桔溃疡病自动识别算法。提出了一种基于特征选择准则的Boosting学习算法,采用对称交叉熵作为弱分类器的相似度评价。将弱分类器相似度与Boosting学习过程相结合学习出更优化的弱分类器,对溃疡病斑图象进行特征选取和学习,建立了自适应的病斑特征模型,最后利用该模型完成溃疡病自动识别。实验结果表明,这种算法避免了Boosting算法进行特征提取时的缺点,减少了选取结果中的冗余,尤其在进行高维特征选取时,能够提高特征选取速度,使选取的特征更具代表性。
To automatically detect citrus canker lesion, a theoretically justfied learning algorithm based on boosting was proposed. Symmetric cross entropy was used as the measure of similarity for weak classifiers. An optimal weak leaner was derived from AdaBoost algorithm. Using this learner, efficient features were selected and an adaptive citrus canker lesion model was constructed. A simulation system based on the model was tested and experiment results show that this method can overcome the disadvantage of boosting algorithm, solving the problem that there is redundancy in the selected features, especially in high-dimension feature selection .And the algorithm is proved to speed feature selection and get more efficient features.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2008年第8期2056-2058,2104,共4页
Journal of System Simulation
基金
教育部博士点基金项目资助(20050611027)
重庆市自然科学基金资助(CSTC2006BB1347)
关键词
特征选择
对称交叉熵
机器视觉
分类器
feature selection
symmetric cross entropy
machine vision
classifier