光电工程
光電工程
광전공정
OPTO-ELECTRONIC ENGINEERING
2014年
9期
45-50
,共6页
高光谱影像分类%维数约简%邻域保持嵌入%半监督学习
高光譜影像分類%維數約簡%鄰域保持嵌入%半鑑督學習
고광보영상분류%유수약간%린역보지감입%반감독학습
hyperspectral image classification%dimension reduction%neighborhood preserving embedding%semi-supervised learning
为进一步提高邻域保持嵌入算法(NPE)在高光谱影像分类中的识别性能，提出一种改进的半监督邻域保持嵌入(SSNPE)算法。首先，该算法在NPE算法的基础上同时利用同类标记样本和邻域未标记样本获得数据的邻域嵌入结构。然后，通过增加近邻标记样本的权重加大降维数据的鉴别性。最后，通过利用k近邻分类器(KNN)对样本进行分类得到该算法在数据集上的分类性能。在 Urban、Indian 高光谱影像数据集上的实验结果表明，改进的算法的分类精度相比其他算法提高了约8.3%、6.2%以上，分类性能上有了较为明显的提高。
為進一步提高鄰域保持嵌入算法(NPE)在高光譜影像分類中的識彆性能，提齣一種改進的半鑑督鄰域保持嵌入(SSNPE)算法。首先，該算法在NPE算法的基礎上同時利用同類標記樣本和鄰域未標記樣本穫得數據的鄰域嵌入結構。然後，通過增加近鄰標記樣本的權重加大降維數據的鑒彆性。最後，通過利用k近鄰分類器(KNN)對樣本進行分類得到該算法在數據集上的分類性能。在 Urban、Indian 高光譜影像數據集上的實驗結果錶明，改進的算法的分類精度相比其他算法提高瞭約8.3%、6.2%以上，分類性能上有瞭較為明顯的提高。
위진일보제고린역보지감입산법(NPE)재고광보영상분류중적식별성능，제출일충개진적반감독린역보지감입(SSNPE)산법。수선，해산법재NPE산법적기출상동시이용동류표기양본화린역미표기양본획득수거적린역감입결구。연후，통과증가근린표기양본적권중가대강유수거적감별성。최후，통과이용k근린분류기(KNN)대양본진행분류득도해산법재수거집상적분류성능。재 Urban、Indian 고광보영상수거집상적실험결과표명，개진적산법적분류정도상비기타산법제고료약8.3%、6.2%이상，분류성능상유료교위명현적제고。
Neighborhood Preserving Embedding (NPE) algorithm is a sub-space learning method, which has the ability to preserve the local neighboring structure information of the date. In order to improve the recognition function of the NPE algorithm used in hyperspectral image classification, we proposed an improved Semi-supervised Neighborhood Preserving Embedding (SSNPE) algorithm. Firstly, the algorithm uses both the labeled samples and the unlabeled samples of the neighborhood to get the neighborhood embedding structure. Secondly, improve the classification feature of the samples through raising weight of the labeled neighboring samples. Finally, get the classification function through using k-nearest Neighboring (KNN) classifier to classify the data set. The experimental results on the Urban, Indian Pine data sets show that the classification rate of the proposed algorithm is improved by more than about 8.3%, 6.2% compared to other algorithms, respectively, and thus the recognition performance has been improved clearly.
