高光谱图像压缩存储

利用SVD函数只保存前k个主成分

from scipy.sparse.linalg import svds
import numpy as np

def svd_image_reconstruction(image, k, return_USV=False):
    C, W, H = image.shape

    # 将图像平铺为 Cx(WH)
    image_reshaped = image.reshape(C, W * H)

    # SVD分解
    # U, S, Vt = svd(image_reshaped, full_matrices=False)
    U, S, Vt = svds(image_reshaped, k=k)

    # 保留前k个主成分
    U_k = U[:, :k]
    S_k = np.diag(S[:k])
    Vt_k = Vt[:k, :]
    U_k, S_k, Vt_k = U_k.astype(np.float16), S_k.astype(np.float16), Vt_k.astype(np.float16)

    reconstructed_image = np.dot(U_k, np.dot(S_k, Vt_k)).reshape(C, W, H)

    if return_USV:
        return reconstructed_image, (U_k, S_k, Vt_k)
    else:
        return reconstructed_image