Skip to content

luckywilliam111/hsipl_algo

BranchesTags

Repository files navigation

Algorithm Package

這是一個由國立雲林科技大學-高光譜訊號與影像處理實驗室成員WEN所開發/整合而成的演算法套件，演算法主要適用於高光譜影像，包括影像前處理與背景抑制演算法，取得方式可以透過 cmd 輸入指令 pip install hsipl-algo==[version] 以取得最新版本套件

image

資料夾階層

hsipl-algo
|---- build
|  |---- bdist.win-amd64
|  |---- lib
|     |---- hsipl_algo
|       |---- Anomaly_Algorithm.py
|       |---- Calc_Band.py
|       |---- CEM_BS_Method.py
|       |---- CNN_BS_Method.py
|       |---- Component_BS_Method.py
|       |---- CT_BS_Method.py
|       |---- Find_EndMember.py
|       |---- get_Threshold.py
|       |---- Hierarchical_Algorithm.py
|       |---- Other_BS_Method.py
|       |---- RPCA_Decomposition.py
|       |---- PCA.py
|       |---- ICA.py
|       |---- MNF.py
|       |---- Target_Algorithm.py
|       |---- Weighted_Target_Algorithm.py
|
|---- dist
|  |---- hsipl_algo-1.2.0.tar
|  |---- hsipl_algo-1.2.0-py3-none-any.whl
|
|---- hsipl_algo
|  |---- Anomaly_Algorithm.py
|  |---- Calc_Band.py
|  |---- CEM_BS_Method.py
|  |---- CNN_BS_Method.py
|  |---- Component_BS_Method.py
|  |---- CT_BS_Method.py
|  |---- Find_EndMember.py
|  |---- get_Threshold.py
|  |---- Hierarchical_Algorithm.py
|  |---- Other_BS_Method.py
|  |---- RPCA_Decomposition.py
|  |---- PCA.py
|  |---- ICA.py
|  |---- MNF.py
|  |---- Target_Algorithm.py
|  |---- Weighted_Target_Algorithm.py
|
|---- hsipl_algo.egg-info
|  |---- dependency_links.txt
|  |---- PKG-INFO
|  |---- SOURCES.txt
|  |---- top_level.txt
|
|---- image
|  |---- install.png
|  |---- apple.jpg
|  |---- GA_Low_Rank_Matrix.png
|  |---- GA_Sparse_Matrix.png
|  |---- Target_Detection.png
|  |---- Anomaly_Detection.png
|  |---- Leather_Image.jpg
|  |---- PPI_EndMember.png
|  |---- N_FINDR_EndMember.png
|  |---- Leather_ROI_Image.png
|  |---- Leather_CEM_Band_Selection.png
|  |---- Leather_Fusion_Band_Selelction_Target.png
|  |---- Leather_Fusion_SF_CTBS_Band_Selelction.png
|  |---- CNN_Model.png
|  |---- Leather_CNN_Entropy_Band_Selection.png
|  |---- Leather_CNN_Variance_Band_Selection.png
|  |---- Leather_PCA_Band_Selection.png
|  |---- Leather_ICA_Band_Selection.png
|
|---- .gitattributes
|---- LICENSE
|---- README.md
|---- setup.py

演算法分類

  • 影像前處理演算法

    • 端元選擇
      • 像素點純度指數(PPI)
      • 自動目標點生成程序(ATGP)
      • 單純形增長算法(SGA)
      • N-FINDR
    • 估計波段選擇數
      • HFC
      • NWHFC
    • 波段選擇
      • 基於CEM-波段選擇
        • 最小能量限制法之波段關聯限制(CEM-BCC)
        • 最小能量限制法之波段關聯(CEM-BCM)
        • 最小能量限制法之波段依賴最小化(CEM-BDM)
      • 基於相關係數-波段選擇
        • 相關係數-波段選擇(BS-Corrcoef)
      • 基於熵-波段選擇
        • 熵-波段選擇(BS-Entropy)
        • 卷積特徵-熵-波段選擇(CNN-Entropy-Band-Selection)
      • 基於標準差-波段選擇
        • 標準差-波段選擇(BS-STD)
      • 基於方差-波段選擇
        • 卷積特徵-方差-波段選擇(CNN-Variance-Band-Selection)
      • 基於約束目標-波段選擇
        • 最小方差-波段優先級
          • MinV-BP
          • FminV-BP
        • 最大方差-波段優先級
          • MaxV-BP
          • BmaxV-BP
        • 循序前向約束目標-波段選擇
          • SF-CTBS
          • SF_TCIMBS
        • 循序後向約束目標-波段選擇
          • SB-CTBS
          • SB-TCIMBS
      • 基於融合波段約束目標-波段選擇
        • 融合波段最小方差-波段優先級(BS-MinV-BP)
        • 融合波段最大方差-波段優先級(BS-MaxV-BP)
        • 融合波段循序前向約束目標-波段選擇(BS-SF-CTBS)
        • 融合波段循序後向約束目標-波段選擇(BS-SB-CTBS)
      • 基於均勻分配-波段選擇
        • 均勻分配-波段選擇(uniform-BS)
      • 基於成分分析法-波段選擇
        • 主成分分析法-波段選擇(PCA-BS)
        • 獨立成分分析法-波段選擇(ICA-BS)
    • 影像分解
      • Robust-PCA-分解法
        • Robust-PCA-Kernel
          • GA-Kernel
          • GM-Kernel
          • Godec-Kernel
          • GreGoDec-Kernel
          • OPRMF-Kernel
          • PCP-Kernel
          • PRMF-Kernel
          • SSGoDec-Kernel
          • SVT-Kernel
          • TGA-Kernel
    • 影像降維
      • Principal Component Analysis (PCA) - 主成分分析法
      • Independent Component Analysis (ICA) - 獨立成分分析法
      • Minimum Noise Fraction (MNF) - 最小噪聲轉換

  • 背景抑制演算法

    • 目標檢測演算法
      • 非權重式檢測演算法
        • 自相關矩陣
          • 最小能量限制法(CEM)
          • 子分割-最小能量限制法(Subset-CEM)
          • 滑動視窗-最小能量限制法(SW-CEM)
          • 自適應滑動視窗-最小能量限制法(SW-CEM)
          • 目標限制干擾最小化濾波器(TCIMF)
          • 自相關濾波器-自適應一致估計法(R-ACE)
          • 自相關濾波器-馬哈拉諾比斯距離(RMD)
          • 自相關濾波器-基於距離之匹配濾波器(RMFD)
        • 共變異數矩陣
          • 匹配濾波器(MF)
          • 自適應匹配濾波器(AMF)
          • 光譜匹配濾波器(AMF)
          • 共變異數-自適應一致估計法(K-ACE)
          • 共變異數-馬哈拉諾比斯距離(KMD)
          • 共變異數-基於距離之匹配濾波器(KMFD)
        • 距離
          • 歐式距離(ED)
        • 光譜角度
          • 光譜角度匹配(SAM)
        • 光譜散度
          • 光譜資訊分散度(SID)
        • 子空間投影
          • 正交子空間投影(OSP)
          • 最小二乘-正交子空間投影(LSOSFP)
          • Kernel最小二乘-正交子空間投影(KLSOSP)
        • SID-SAM混和
          • SID-tan(SAM)
          • SID-sin(SAM)
        • 未分類
          • CBD
          • JMD
          • OPD
          • TD
          • 廣義相似比測試(GLRT)
      • 權重式檢測演算法
        • 權重式-背景抑制
          • 歐式距離-權重式最小能量限制法(ED-CEM)
          • 光譜資訊分散度-權重式最小能量限制法(SID-CEM)
        • 迭代權重式-背景抑制
          • 自相關矩陣
            • 分層抑制式-最小能量限制法(hCEM)
          • 共變異數矩陣
            • 分層抑制式-共變異數-自適應一致估計法(hACE)
            • 分層抑制式-共變異數-馬哈拉諾比斯距離(hKMD)
            • 分層抑制式-自適應匹配濾波器(hAMF)
    • 異常檢測演算法
      • 自相關矩陣
        • 自相關濾波器-Reed-Xiaoli檢測器(R-RXD)
        • 滑動視窗-Reed-Xiaoli檢測器((SW-RXD)
        • 低概率目標檢測器(LPTD)
      • 共變異數矩陣
        • 共變異數-Reed-Xiaoli檢測器(K-RXD)
        • 匹配自相關濾波器-Reed-Xiaoli檢測器(M-RXD)
        • 正規化-RXD(N-RXD)
        • 統一目標檢測器(UTD)
        • 統一目標自相關濾波器-Reed-Xiaoli檢測器(UTD-RXD)

Robust PCA Decomposition Example Code

import numpy as np
import matplotlib.pyplot as plt
import hsipl_algo.RPCA_Decomposition as hRD

data = plt.imread('apple.jpg')

plt.figure()
plt.imshow(data)
plt.show()

data = np.double(data)

x, y, z = data.shape

re_data = data.reshape(x * y, z)

L, S = hRD.GA(re_data)

L = L.reshape(x, y, z)

S = S.reshape(x, y, z)

plt.figure()
plt.imshow(L / 255)
plt.show()

plt.figure()
plt.imshow(S / 255)
plt.show()

drawingdrawingdrawing

Principal Component Analysis Reduce Dimension (2021/5/22 Update)

import numpy as np
import matplotlib.pyplot as plt
from hsipl_algo.PCA import PCA 

im = plt.imread('pic1.tif')
im = np.double(im) / 255.0

x, y, z = im.shape

im = im.reshape(x * y, z)

pca = PCA(n_components=2)

pca.fit(im)

tran = pca.transform(im)

explained_variance_ratio = pca.explained_variance_ratio_

components = pca.components_

plt.imshow(tran[:, 0].reshape(1000, 1300))

Independent Component Analysis Reduce Dimension (2021/5/22 Update)

import numpy as np
import matplotlib.pyplot as plt
from hsipl_algo.ICA import ICA 

im = plt.imread('pic1.tif')
im = np.double(im) / 255.0

x, y, z = im.shape

im = im.reshape(x * y, z)

ica = ICA(n_components=2)

ica.fit(im)

tran = ica.transform(im)

components = ica.components_

mixing = ica.mixing_

plt.imshow(tran[:, 0].reshape(1000, 1300))

Minimum Noise Fraction Reduce Dimension And Denoise (2021/5/22 Update)

import numpy as np
import matplotlib.pyplot as plt
from hsipl_algo.MNF import calc_stats, noise_from_diffs, mnf

im = plt.imread('pic1.tif')
im = np.double(im) / 255.0

x, y, z = im.shape

signal = calc_stats(im)
noise = noise_from_diffs(im)
mnfr = mnf(signal, noise)

denoised = mnfr.denoise(im, snr=10)
reduced = mnfr.reduce(im, snr=10)
reduced = mnfr.reduce(im, num=50)

Find End-Member Example Code

import numpy as np
import scipy.io as sio
import matplotlib.pyplot as plt
import hsipl_algo.Find_EndMember as hFE

data = sio.loadmat('A_6_mnf.mat')
data = data['im_mnf']

data = np.double(data)

bs = sio.loadmat('A_6_CEM_SF_CTBS_band_select_result.mat')
bs = (bs['SF_CTBS_band_select_result']).transpose()
bs = bs.reshape(bs.shape[0])

bs_data = data[:, :, bs]

plt.figure()
plt.imshow(data[:, :, 100], cmap='gray')
plt.show()

x, y, z = data.shape

xx, yy, score, result = hFE.PPI(bs_data, 1500)

score = score.reshape(x, y)

coordinate = []

plt.figure()
plt.imshow(data[:, :, 100], cmap='gray')
for i in range(5):
    ppi_score = xx[0, -1-i]
    x_y = np.argwhere(score == ppi_score)
    
    try:
        for j in range(x_y.shape[0]):
            coordinate.append([ppi_score, x_y[j, 0], x_y[j, 1]])
            plt.plot(np.int(x_y[j, 1]), np.int(x_y[j, 0]), '.r')
            plt.text(np.int(x_y[j, 1])-7, np.int(x_y[j, 0])-7, 'PPI Score= ' + str(np.int(ppi_score)), fontsize=10, color='blue', horizontalalignment='right', verticalalignment='bottom', bbox=dict(facecolor='green', alpha=0.3))
    except Exception:
        pass
    
plt.show()

endmemberindex = hFE.N_FINDR(bs_data, 4)

plt.figure()
plt.imshow(data[:, :, 100], cmap='gray')
for i in range(endmemberindex.shape[0]):
    plt.plot(np.int(endmemberindex[i, 0]), np.int(endmemberindex[i, 1]), '.r')
    plt.text(np.int(endmemberindex[i, 0])-7, np.int(endmemberindex[i, 1])-7, 'N-FINDR EndMember= ' + str(np.int(i+1)), fontsize=10, color='blue', horizontalalignment='right', verticalalignment='bottom', bbox=dict(facecolor='green', alpha=0.3))
plt.show()

drawingdrawingdrawing

CEM / CTBS / Fusion-CTBS Band-Selection Example Code

import numpy as np
import scipy.io as sio
import matplotlib.pyplot as plt
import hsipl_algo.CEM_BS_Method as hCM
import hsipl_algo.CT_BS_Method as hCBM

data = sio.loadmat('A_6_mnf.mat')
data = data['im_mnf']

d = sio.loadmat('A_6_d.mat')
d = d['d']

coordinate = sio.loadmat('A_6_coordinate.mat')
coordinate = coordinate['coordinate']

data = np.double(data)

x, y, z = data.shape

plt.figure()
plt.imshow(data[:, :, 100], cmap='gray')
plt.show()

data_roi = data[213:263, 170:212, :]

plt.figure()
plt.imshow(data_roi[:, :, 100], cmap='gray')
plt.show()

bs_cem = hCM.CEM_BDM(data_roi[:, :, 30:200], 5)
bs_cem = bs_cem + 30

plt.figure()
plt.plot(d, 'b', label='d')
for i in range(bs_cem.shape[0]):
    plt.axvline(x = bs_cem[i], color='y', linestyle='--')
plt.legend()
plt.show()

bs_ctbs = hCBM.SF_CTBS(data[:, :, 30:200], d[30:200], 5)
bs_ctbs = bs_ctbs + 30

plt.figure()
plt.plot(d, 'b', label='d')
for i in range(bs_ctbs.shape[0]):
    plt.axvline(x = bs_ctbs[i], color='y', linestyle='--')
plt.legend()
plt.show()

d1 = data[coordinate[0, 1], coordinate[0, 0], :].reshape(z, 1)
d2 = data[coordinate[1, 1], coordinate[1, 0], :].reshape(z, 1)
d3 = data[coordinate[2, 1], coordinate[2, 0], :].reshape(z, 1)

plt.figure()
plt.imshow(data[:, :, 100], cmap='gray')
plt.plot(coordinate[0, 0], coordinate[0, 1], '.r')
plt.text(coordinate[0, 0]-7, coordinate[0, 1]-7, 'd1', fontsize=10, color='blue', horizontalalignment='right', verticalalignment='bottom', bbox=dict(facecolor='green', alpha=0.3))
plt.plot(coordinate[1, 0], coordinate[1, 1], '.g')
plt.text(coordinate[1, 0]-7, coordinate[1, 1]-7, 'd2', fontsize=10, color='blue', horizontalalignment='right', verticalalignment='bottom', bbox=dict(facecolor='green', alpha=0.3))
plt.plot(coordinate[2, 0], coordinate[2, 1], '.b')
plt.text(coordinate[2, 0]-7, coordinate[2, 1]-7, 'd3', fontsize=10, color='blue', horizontalalignment='right', verticalalignment='bottom', bbox=dict(facecolor='green', alpha=0.3))
plt.show()

d = np.hstack([d1, d2, d3])

bs_fusion_ctbs = hCBM.BS_SF_CTBS(data[:, :, 30:200], d[30:200, :], 5)
bs_fusion_ctbs = bs_fusion_ctbs + 30

color = ['red', 'green', 'blue']

plt.figure()
for i in range(bs_fusion_ctbs.shape[0]):
    if i < d.shape[1]:
        plt.plot(d[:, i], color=color[i], label='d' + str(i+1))
    plt.axvline(x = bs_fusion_ctbs[i], color='y', linestyle='--')
plt.legend()
plt.show()

drawingdrawing drawingdrawing drawingdrawing

CNN Feature Entropy / Variance Band-Selection Example Code (2021/3/7 update)

import scipy.io as sio
import matplotlib.pyplot as plt
import hsipl_algo.CNN_BS_Method as hCNNBM

#================================ Load Data ===================================

path = 'algo_test/data1/'

data = sio.loadmat(path + 'A_6_mnf.mat')
data = data['im_mnf']

d = sio.loadmat(path + 'A_6_d.mat')
d = d['d']

HIM = data[:, :, 30:200]

x, y, z = HIM.shape

re_HIM = HIM.reshape(z, x, y, 1)

num_band = 5

#==============================================================================
#================================= Create Model ===============================

net = []

net.append({'name': 'input',
            'layer': 'Input',
            'h': re_HIM.shape[1],
            'w': re_HIM.shape[2],
            'channel': re_HIM.shape[3]})

net.append({'name': 'c1',
            'layer': 'Conv2D',
            'filters': 16,
            'kernel_size': 3,
            'strides': 1,
            'padding': 'same',
            'activation': 'relu'})

net.append({'name': 'bn1',
            'layer': 'BatchNormalization'})

net.append({'name': 'm1',
            'layer': 'MaxPooling2D',
            'pool_size': 2,
            'strides': None,
            'padding': 'same'})

net.append({'name': 'drop1',
            'layer': 'Dropout',
            'rate': 0.1})

net.append({'name': 'c2',
            'layer': 'Conv2D',
            'filters': 8,
            'kernel_size': 3,
            'strides': 1,
            'padding': 'same',
            'activation': 'relu'})

net.append({'name': 'bn2',
            'layer': 'BatchNormalization'})

net.append({'name': 'm2',
            'layer': 'MaxPooling2D',
            'pool_size': 2,
            'strides': None,
            'padding': 'same'})

net.append({'name': 'f1',
            'layer': 'Flatten'})

net.append({'name': 'd1',
            'layer': 'Dense',
            'units': 4096,
            'activation': 'softmax'})

model = hCNNBM.cnn_Featur_Model(net)

#==============================================================================
#========================== Get Feature Map Get Band ==========================

band_select_entropy = hCNNBM.CNN_Entropy_Band_Selection(HIM, model, num_band)
band_select_entropy = band_select_entropy + 30

band_select_variance = hCNNBM.CNN_Variance_Band_Selection(HIM, model, num_band)
band_select_variance = band_select_variance + 30

#==============================================================================
#================================= Plot Band ==================================

plt.figure()
plt.plot(d, 'r')
for i in range(num_band):
    plt.axvline(x=band_select_entropy[i], color='yellow')
    
plt.show()

plt.figure()
plt.plot(d, 'r')
for i in range(num_band):
    plt.axvline(x=band_select_variance[i], color='yellow')
    
plt.show()

#==============================================================================

CNN-Model

drawing

drawingdrawingdrawing

Principal component analysis / Independent Component Analysis Band-Selection Example Code

import scipy.io as sio
import matplotlib.pyplot as plt
import hsipl_algo.Component_BS_Method as hCBM

#================================ Load Data ===================================

path = 'algo_test/data1/'

data = sio.loadmat(path + 'A_6_mnf.mat')
data = data['im_mnf']

d = sio.loadmat(path + 'A_6_d.mat')
d = d['d']

HIM = data[:, :, 30:200]

num_band = 5

n_components = 30

#==============================================================================

#========================= PCA / ICA Band-Selection ===========================

band_select_pca = hCBM.PCA_BS(HIM, num_band)
band_select_pca = band_select_pca + 30

band_select_ica = hCBM.ICA_BS(HIM, n_components, num_band)
band_select_ica = band_select_ica + 30

#==============================================================================

#================================= Plot Band ==================================

plt.figure()
plt.plot(d, 'r')
for i in range(num_band):
    plt.axvline(x=band_select_pca[i], color='yellow')
    
plt.show()

plt.figure()
plt.plot(d, 'r')
for i in range(num_band):
    plt.axvline(x=band_select_ica[i], color='yellow')
    
plt.show()

#==============================================================================

drawingdrawingdrawing

Target / Anomaly Detection Example Code

import numpy as np
import matplotlib.pyplot as plt
import hsipl_algo.Target_Algorithm as hTA
import hsipl_algo.Anomaly_Algorithm as hAA

data = plt.imread('apple.jpg')

plt.figure()
plt.imshow(data)
plt.show()

data = np.double(data)

x, y, z = data.shape

d = data[70, 150, :].reshape(z, 1)

result = hTA.CEM(data, d)

plt.figure()
plt.imshow(result, cmap='gray')
plt.show()

result = hAA.R_RXD(data)

plt.figure()
plt.imshow(result, cmap='gray')
plt.show()

drawingdrawingdrawing

WEN-Github

About

No description, website, or topics provided.

Resources

Readme

License

View license
Activity

Stars

0 stars

Watchers

1 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

No packages published

Languages