【摘要】：基于BP-Adaboost的目標(biāo)分類算法用于雷達目標(biāo)分類具有良好的效果.隨著訓(xùn)練樣本以及測試樣本數(shù)增加,經(jīng)典"一對多(One vs.Rest,OvR)"BP-Adaboost算法所需用時也隨之增加.提出一種改進的多分類BP-Adaboost算法應(yīng)用在雷達多目標(biāo)分類上,在提高分類準(zhǔn)確率的同時,有效地解決經(jīng)典算法在多分類上時間開銷過大的問題.該方法采用二進制方法重新表示樣本數(shù)據(jù)類別,使用Adaboost算法將多個BP神經(jīng)網(wǎng)絡(luò)弱分類器集成起來學(xué)習(xí),通過修改經(jīng)典算法中的損失函數(shù)連續(xù)調(diào)整訓(xùn)練樣本分布和弱分類器的權(quán)重,最終形成一個強分類器.對雷達高分辨率距離像(High Resolution Range Profile,HRRP)數(shù)據(jù)集進行分類仿真結(jié)果表明,相比于單個BP神經(jīng)網(wǎng)絡(luò)基學(xué)習(xí)器,所提算法的分類準(zhǔn)確率提高了5%~10%,相比于經(jīng)典的"一對多"BP-Adaboost算法,該算法所需用時僅為傳統(tǒng)算法的1/2~1/3.
[Abstract]:The target classification algorithm based on BP-Adaboost has good effect in radar target classification. With the increase of training samples and the number of test samples, the time required for the classical "one-to-many (One vs.Rest,OvR)" BP-Adaboost algorithm also increases. An improved multi-classification BP-Adaboost algorithm is proposed for radar multi-target classification, which not only improves the classification accuracy, but also effectively solves the problem that the classical algorithm has too much time overhead in multi-classification. In this method, the binary method is used to rerepresent the sample data category, and the Adaboost algorithm is used to integrate multiple BP neural network weak classifiers. By modifying the loss function in the classical algorithm, the training sample distribution and the weight of the weak classifiers are continuously adjusted, and finally a strong classifier is formed. The classification simulation results of radar high resolution range profile (High Resolution Range Profile,HRRP) data set show that compared with a single BP neural network based learner, the classification accuracy of the proposed algorithm is improved by 5% 鈮，

本文編號：2526330