基于Fisher线性鉴别特征融合的海底底质分类研究

Seabed Classification Approach Based on Fisher Linear Discriminant Feature Fusion

提出了一种基于Fisher线性鉴别特征融合的海底底质分类方法。首先,分析了声学遥感可以进行海底底质分类与识别,但是现有的分类方法,存在特征量维数较大、分类器设计复杂、分类效果不佳、不能获得最佳鉴别矢量等缺陷。因此,提出一种基于Fisher线性鉴别特征融合的海底底质分类方法。该方法首先提取同一样本的12个统计特征量,然后利用特征融合技术将这12个特征量组合在一起,构成一个新的特征矢量空间,最后在该空间中利用Fisher线性鉴别分析进行最优鉴别特征提取。其次,以岩石、砾石、沙、泥四种沉积物为实验对象来开展水下实验。对回波数据进行预处理,然后对每一样本提取12个统计特征量,采用串行融合方法进行特征组合,最后采用Fisher线性判别分析得到最佳鉴别矢量特征,并送入最近邻分类器进行分类。最后,通过大量的实验数据对比,发现基于Fisher线性鉴别特征融合的海底底质分类方法比PCA方法和单一特征方法具有更高的正确分类率。

The paper proposed a seabed sediments classification method based on Fisher linear discriminant feature fusion.Although acoustic remote sensing method can be used in classification and recognition of seabed sediments,it goes along with deficiency oflarge dimension,complicated classifier design,poor classification effect and incapabilityof obtaining the best discriminant vector.The proposed method firstly extracted 12 statistical features of one sample,and then combined the 12 features together with the feature fusion technology,forming a new feature vector space.The optimal discriminant feature can be extracted in the feature vector space by using Fisher linear discriminant.Next,underwater experiments are carried out with four kinds of sediments,rock,gravel,sand,and mud.The echo data are preprocessed,and then 12 statistical features are extracted for each sample.Feature combination is achieved by serial fusion and Fisher discriminant analysis is applied to get the best discriminant vector features.The features are sent to nearest neighbor classifier for classification.Alarge number of experimental data comparison shows that the seabed sediments classification method based on Fisher linear discriminant feature fusion has higher correct classification rate than PCA method and single feature method.