Research And Application Of AdaBoost Based On Fitting Weak Classifier

Adaboost algorithm is currently a more popular machine learning algorithm.It is widely used in face detection and other fields by constructing a series of weak classifiers and combining them into strong classifiers.For the current AdaBoost algorithm's training time is too long to converge,detection accuracy is poor,and noise sensitivity,etc.,this paper designs a polynomial fitting weak classifier from the perspective of feature distribution fitting.The method of fitting optimization is used to reduce the weak The classifier also improves detection accuracy.From the perspective of fuzzy comprehensive evaluation,a weak classifier for fuzzy evaluation is designed,which effectively improves the robustness of the strong classifier.The improved algorithm is applied to face detection,which improves its detection performance.The main research work includes:First,in order to solve the problem of poor accuracy caused by selecting weak classifiers by minimizing the training error rate,and the problem that single thresholds are slow to train as weak classifiers,it is difficult to converge.AdaBoost algorithm.First,for each feature,a mapping relationship is established between the feature value and the labeled value,and the least square method is introduced to solve the fitting polynomial function and converted into discrete classification values to obtain a weak classifier.Secondly,from the obtained many weak classifiers,the weak classifier with the smallest classification error is selected as the best weak classifier in this round of iteration to form a new AdaBoost strong classifier.The algorithm greatly reduces the number of weak classifiers to be selected,effectively reduces the time complexity of sample training,and improves the detection accuracy.Secondly,in order to optimize the phenomenon that AdaBoost algorithm is sensitive to noise data and affect its classification performance during training,an AdaBoost algorithm based on fuzzy comprehensive evaluation is proposed.First,the eigenvalue distribution curve of each feature in the training sample set is divided into a plurality of Gaussian unimodal functions,and each Gaussian unimodal function corresponds to a fuzzy membership function to discretize a single-factor matrix set.Then build a fuzzy comprehensive evaluation model based on a single factor matrix,change the evaluation index weight vector for search and optimization,and select the fuzzy evaluation weak classifier composed of the weight vector with the smallest total error and the single factor matrix set as the best weak classification for the current iteration.,The combination constitutes a new AdaBoost strong classifier.The algorithm can accept the variation of sample feature values within a certain range without changing the label value,and has better fault tolerance for learning noise samples,thereby effectively improving the robustness of the classifier.Finally,the AdaBoost face detection algorithm was used to design a face detection system on the embedded platform.In order to solve the problem of possible false detection and miss detection of the face detection system in actual complex scenes,combined with the optimized AdaBoost algorithm to improve,The optimization of the improved algorithm in real-time face detection accuracy and anti-noise performance is studied.In summary,based on the existing AdaBoost face detection algorithms,this paper studies the AdaBoost algorithm based on a polynomial fitting weak classifier and the fuzzy comprehensive evaluation weak classifier,and applies the algorithm to face detection.The application verifies the performance improvement of the algorithm.