cls.py 文件源码

def SID_classifier(M, E, threshold):
    """
    Classify a HSI cube M using the spectral information divergence
    and a spectral library E.
    This function is part of the NormXCorr class.

    Parameters
        M : numpy array
          a HSI cube ((m*n) x p)

        E : numpy array
          a spectral library (N x p)

    Returns : numpy array
          a class map ((m*n))
    """
    def prob_vector_array(m):
        pv_array = np.ndarray(shape=m.shape, dtype=np.float32)
        sum_m = np.sum(m, axis=1)
        pv_array[:] = (m.T / sum_m).T
        return pv_array + np.spacing(1)

    mn = M.shape[0]
    N = E.shape[0]
    p = prob_vector_array(M)
    q = prob_vector_array(E)
    sid = np.ndarray((mn, N), dtype=np.float)
    for i in range(mn):
        pq = q[0:,:] * np.log(q[0:,:] / p[i,:])
        pp = p[i,:] * np.log(p[i,:] / q[0:,:])
        sid[i,:] = np.sum(pp[0:,:] + pq[0:,:], axis=1)
    if isinstance(threshold, float):
        cmap = _single_value_min(sid, threshold)
    elif isinstance(threshold, list):
        cmap = _multiple_values_min(sid, threshold)
    else:
        return np.argmin(sid, axis=1), sid
    return cmap, sid