遙降維（DR）是一種旨在減少維數據還原為低維表示形式以便能夠分析大量多元數據的方法。DR方法對於提取產生大量光譜信息並避免出現維數問題的詛咒的高光譜圖像的重要特徵至關重要。諸如流形學習之類的DR方法方法，嘗試保留樣本之間的局部結構。諸如非線性方法之類的另一種方法試圖克服面對非線性分佈數據時將數據投影到低維子空間中的線性限制。這項研究提出了多核最近特徵線嵌入（MKFLE）用於高光譜圖像分類。MKL通過使用許多內核的線性組合來生成用於判別分析的非線性特徵空間。NFLE方法通過判別分析將NFL嵌入到變換中，旨在保留希爾伯特空間中樣本的局部結構。MKFLE的組合旨在克服噪聲變異和高度非線性的數據分佈，並在高光譜圖像分類中獲得更準確和穩定的結果。實驗是使用三種不同的高光譜數據集進行的，即印度松樹數據集，帕維亞市中心數據集和帕維亞大學數據集。結果表明，該方法比單核方法具有更穩定的分類結果。;Dimensionality reduction (DR) is a method that aims to reduce the high-dimensional data into lower-dimensional representation to be able to analyze a large amount of multivariate data. DR method is essential to extract important features of hyperspectral images that generate abundant spectral information and avoid the curse of dimensionality problem. DR method approach such as manifold learning, try to preserve the local structure among samples. Another approach such as nonlinear method, try to overcome the linear limitations in projecting data into the lower-dimensional subspace when faced with nonlinearly distributed data. This study presents multiple kernel feature line embedding (MKFLE) for hyperspectral image classification. MKL generates a non-linear feature space for discriminant analysis by using a linear combination of many kernels. NFLE method embedded the NFL into the transformation through discriminant analysis and aim to preserve the local structure of samples in the Hilbert space. The combination of MKFLE aims to overcome noise variation and high-degree non-linear data distribution and get a more accurate and stable result in hyperspectral image classification. Experiments were carried out using three different Hyperspectral datasets namely Indian Pines, Pavia City Center, and Pavia University. The result shows that the proposed approach has a more stable result of classification compared to the single kernel approach.
