This paper presents a technique for hyperspectral image classification and dimensionality reduction using orthogonal subspace projection (OSP). The method aims to reduce data dimensionality, suppress unwanted spectral signatures, and detect the presence of specific spectral signatures. The core concept involves projecting each pixel vector onto a subspace orthogonal to the unwanted signatures, which optimizes interference suppression in the least squares sense. After removing interference, the projection onto the desired signature maximizes the signal-to-noise ratio, resulting in a single component image that represents the classification for the desired signature. The OSP operator can be extended to multiple signatures, reducing the dimensionality of \( k \) signatures and classifying the hyperspectral image simultaneously. The approach is applicable to both spectrally pure and mixed pixels. The paper is structured into several sections: Problem Formulation, Hyper spectral Pixel Classification, Simulation Results, and Experimental Results Using AVIRIS Data. The problem formulation section defines the hyperspectral image cube and the linear model for mixed pixels. The Hyper spectral Pixel Classification section details the OSP classification operator, which first eliminates interfering signatures and then maximizes the signal-to-noise ratio for the desired signature. The simulation results demonstrate the effectiveness of the OSP technique in detecting signatures at low abundance levels and signal-to-noise ratios. The experimental results using AVIRIS data from the Lunar Crater Volcanic Field show that the OSP technique produces component images that align with known geological attributes of the scene. The authors conclude that the OSP technique effectively reduces hyperspectral data dimensionality and accurately detects hyperspectral signatures, making it a valuable tool for various remote sensing applications.This paper presents a technique for hyperspectral image classification and dimensionality reduction using orthogonal subspace projection (OSP). The method aims to reduce data dimensionality, suppress unwanted spectral signatures, and detect the presence of specific spectral signatures. The core concept involves projecting each pixel vector onto a subspace orthogonal to the unwanted signatures, which optimizes interference suppression in the least squares sense. After removing interference, the projection onto the desired signature maximizes the signal-to-noise ratio, resulting in a single component image that represents the classification for the desired signature. The OSP operator can be extended to multiple signatures, reducing the dimensionality of \( k \) signatures and classifying the hyperspectral image simultaneously. The approach is applicable to both spectrally pure and mixed pixels. The paper is structured into several sections: Problem Formulation, Hyper spectral Pixel Classification, Simulation Results, and Experimental Results Using AVIRIS Data. The problem formulation section defines the hyperspectral image cube and the linear model for mixed pixels. The Hyper spectral Pixel Classification section details the OSP classification operator, which first eliminates interfering signatures and then maximizes the signal-to-noise ratio for the desired signature. The simulation results demonstrate the effectiveness of the OSP technique in detecting signatures at low abundance levels and signal-to-noise ratios. The experimental results using AVIRIS data from the Lunar Crater Volcanic Field show that the OSP technique produces component images that align with known geological attributes of the scene. The authors conclude that the OSP technique effectively reduces hyperspectral data dimensionality and accurately detects hyperspectral signatures, making it a valuable tool for various remote sensing applications.