Source code for libpyhat.emi.emi

import numpy as np
from pysptools.eea import PPI, FIPPI, NFINDR, ATGP
from spectral.algorithms import smacc




class SMACC:
    def __init__(self):
        self.max_residual_norm = float("Inf")



    def extract(self, spectra, n_endmembers):
        spect_scaled = spectra / np.max(
            spectra
        )  # smacc documentation suggests scaling large values
        # for numerical stability
        self.S, self.F, self.R = smacc(
            np.squeeze(spect_scaled),
            min_endmembers=n_endmembers,
            max_residual_norm=float("Inf"),
        )






def emi(data, col="wvl", emi_method="FIPPI", n_endmembers=6):
    supported_methods = ("FIPPI", "PPI", "N-FINDR", "ATGP", "SMACC")
    try:
        if emi_method.upper() in supported_methods:
            if emi_method == "PPI":
                method = PPI()
            if emi_method == "FIPPI":
                method = FIPPI()
            if emi_method == "N-FINDR":
                method = NFINDR()
            if emi_method == "ATGP":
                method = ATGP()
            if emi_method == "SMACC":
                method = SMACC()

        else:
            print(
                f"{emi_method} is not a supported method.  Supported methods "
                f"are {supported_methods}"
            )
            return 1
    except KeyError:
        print(f"Unable to instantiate class from {emi_method}.")
        return 1

    spectra = data[col].to_numpy()
    if len(spectra.shape) == 2:
        spectra = np.expand_dims(spectra, 0)

    method.extract(spectra, n_endmembers)
    if emi_method == "SMACC":
        endmember_indices = np.any(method.F == 1, axis=1)
    else:
        endmember_indices = method.idx

    indices = np.zeros(spectra.shape[1], dtype=int)
    indices[endmember_indices] = 1
    data[("endmembers", emi_method)] = indices
    return data, endmember_indices