基于AdaBoost算法的级联分类器对绿色荔枝的快速检测方法

Fast Detection Method for Green Litchi by Cascaded Classifier Based on AdaBoost Algorithm

针对绿色荔枝与树叶颜色相似,采摘机器人在自然环境下准确识别较为困难的问题,提出一种基于AdaBoost算法的级联分类器快速检测方法。首先提取MB-LBP特征,并基于积分图技术快速计算其特征值;然后利用AdaBoost算法从MB-LBP特征中构造若干个最优弱分类器,并加权组合成强分类器;最后通过若干个强分类器的级联来构造级联分类器,可获得基于MB-LBP特征的AdaBoost级联分类器。试验表明:该方法对绿色荔枝的识别准确率为92.7%,召回率为81.3%;测试图像的平均处理时间为1.276 s。

Due to the similar color of green litchi and leaves,it is more difficult to accurately identify green litchi in the wild environment.In this regard,this paper proposes a fast detection method for green litchi by cascaded classifier based on AdaBoost algorithm.Firstly,the MB-LBP features are extracted,and their eigenvalues are quickly calculated based on the integral graph technique.Then,several optimal weak classifiers are constructed from the MB-LBP features by AdaBoost algorithm,and weighted into strong classifiers.Finally,the AdaBoost cascade classifier based on MB-LBP feature can be obtained by constructing a cascade classifier by cascading several strong classifiers.Experiments show that the AdaBoost cascade classifier based on MB-LBP features the recognition accuracy of green litchi is 92.7%,the recall rate is 81.3%,and the average processing time of test images is 1.276s.Therefore,the robustness and real-time performance of the algorithm is good,and it also provides a feasible method for green fruit detection.