Classification using spatial shrunken centroids.
使用空間收縮質(zhì)心進(jìn)行分類。
本節(jié)演示了我們在 Cardinal 中引入的空間收縮質(zhì)心方法進(jìn)行統(tǒng)計分析的空間收縮質(zhì)心方法。在這種方法中，我們通過空間平滑調(diào)整最近的收縮質(zhì)心分類器 [2]。該方法使用統(tǒng)計正則化將每個條件的平均譜向全局平均譜縮小。這種收縮允許重要質(zhì)量的自動特征選擇。然后它通過將像素的質(zhì)譜與每個條件下縮小的平均光譜進(jìn)行比較來對像素進(jìn)行分類。空間平滑使用從空間感知聚類 [3] 改編的權(quán)重，包括高斯權(quán)重，以及試圖解釋局部結(jié)構(gòu)的自適應(yīng)權(quán)重。
This section demonstrates the spatial shrunken centroids classification method for statistical analysis we introduce in Cardinal in the spatial Shrunken Centroids method. In this method, we adapt the nearest shrunken centroids classifier [2] with spatial smoothing. This method? uses statistical regularization to shrink each condition's mean spectrum toward the global mean spectrum. This shrinkage allows automated feature selection of important masses. It then classifies pixels by comparing their mass spectra to the shrunken mean spectra of each conditions. The spatial smoothing uses weights adapted from spatially-aware clustering [3], including Gaussian weights, and adaptive weights that attempt to account for local structure.
The parameters to be explicitly provided in the spatialShrunkenCentroids method are:在 spatialShrunkenCentroids 方法中要明確提供的參數(shù)是：
? r: The neighborhood smoothing radius??r：鄰域平滑半徑
? s: The shrinkage parameter??s: 收縮參數(shù)
The s parameter is the shrinkage parameter that enforces sparsity. As s increases, fewer mass features (m=z values) will be used by the classifier, and only the informative mass features will be retained.s 參數(shù)是強(qiáng)制稀疏的收縮參數(shù)。隨著 s 的增加，分類器將使用更少的質(zhì)量特征（m=z 值），并且只會保留信息豐富的質(zhì)量特征。
For a detailed explanation of the shrinkage parameter s, see [2] and [4].有關(guān)收縮參數(shù) s 的詳細(xì)說明，請參見 [2] 和 [4]。
Clustering can also be performed if no response variable y is given, by providing an additional parameter k for the initial number of clusters. See the clustering work below for details.如果沒有給出響應(yīng)變量 y，也可以通過為初始聚類數(shù)提供附加參數(shù) k 來執(zhí)行聚類。有關(guān)詳細(xì)信息，請參閱下面的聚類工作。
2.5.1 Cross-validation?2.5.1 交叉驗證
