一种基于Adaboost.M1的车型分类算法

A Vehicle Classification Algorithm Based on Adaboost.M1

神经网络分类器存在容易出现过学习、欠学习、陷入维数灾以及局部最小等问题,支持向量机分类器也存在运算比较复杂,模型选择和核函数的构造比较困难的问题,而贝叶斯分类器只有在训练样本数趋于无穷时,训练结果才趋于真实的模型,因此,提出了一种基于Adaboost.M1理论的车型分类算法,该算法简单易用,只需要寻找一个精度比随机预测略高的弱分类器,不需要调节任何参数,不需要先验知识,而且有足够的理论支持.最后通过实验验证了该算法进行车型分类的有效性.

Neural network classifiers have problems of over learning,less learning,fall into curse of dimensionality or local minimum,and support vector machine classifiers also have problems of more complex operations,model selection and construction of kernel function is more difficult,and Bayesian classification only in the number of training samples tends to infinity,the training results of the model tends to true.This paper presents a vehicle classification algorithm based on Adaboost.M1.The algorithm is simple to use,and just need to find a weak classifier which′s precision slightly higher than the random prediction,without adjusting any parameters,no prior knowledge,and there is sufficient theoretical support.Finally,experimental results demonstrate the effectiveness of the vehicle classification algorithm based on Adaboost.M1.