基于高光谱图像像元的特殊性质和循环神经网络结合一维卷积网络配合链式分类器实现解混python源码.zip

import numpy as np from keras.models import Sequential, Model from keras.layers import Dense, Dropout, Flatten, Reshape, BatchNormalization, concatenate, Input from keras.layers import Conv2D, MaxPooling2D, SeparableConv2D, ZeroPadding2D, Conv1D, MaxPooling1D from keras.layers import LSTM, Bidirectional, TimeDistributed, RepeatVector, GRU from keras.optimizers import Adam, SGD, Nadam from keras.layers import LeakyReLU from keras.regularizers import l1, l2 from keras.callbacks import TensorBoard, EarlyStopping, ModelCheckpoint import scipy.io as sco import os os.environ["CUDA_VISIBLE_DEVICES"] = '0' #随机对原始数据进行采样生成训练集和测试集 def sample(TrainRate, Data, Label): n, m = np.shape(Data) DataIndex = list(range(m)) TrainData = [] TrainLabel = [] TestData = [] TestLabel = [] for i in range(int(TrainRate*m)): Randindex = int(np.random.uniform(0, len(DataIndex))) TrainData.append(Data[:, Randindex]) TrainLabel.append(Label[:, Randindex]) del (DataIndex[Randindex]) for i in DataIndex: TestData.append(Data[:, i]) TestLabel.append(Label[:, i]) TrainLabel = np.array(TrainLabel) TestLabel = np.array(TestLabel) TrainData = np.array(TrainData) TestData = np.array(TestData) return TrainData, TrainLabel, TestData, TestLabel def tenfold(Data, Label, iter): n, m = np.shape(Data) Data = list(np.transpose(Data)) Label = list(np.transpose(Label)) m_jr = 10000 m_ur = m - m_jr m_jr = m_jr // 10 m_ur = m_ur // 10 TestData = Data[((iter - 1) * m_jr) : (iter * m_jr)] + Data[((iter - 1) * m_ur + m_jr) : (iter * m_ur + m_jr)] TestLabel = Label[((iter - 1) * m_jr) : (iter * m_jr)] + Label[((iter - 1) * m_ur + m_jr) : (iter * m_ur + m_jr)] TrainData = Data[: ((iter - 1) * m_jr)] + Data[(iter * m_jr):((iter - 1) * m_ur + m_jr)] + Data[(iter * m_ur + m_jr):] TrainLabel = Label[: ((iter - 1) * m_jr)] + Data[(iter * m_jr):((iter - 1) * m_ur + m_jr)] + Data[(iter * m_ur + m_jr):] TrainLabel = np.array(TrainLabel) TrainData = np.array(TrainData) TestData = np.array(TestData) TestLabel = np.array(TestLabel) return TrainData, TrainLabel, TestData, TestLabel def mix_sample(TrainRate, Data, Label): n, m = np.shape(Data) m_jr = 10000 m_ur = m - m_jr #对Jasper随机采样 DataIndex = list(range(m_jr)) TrainData = [] TrainLabel = [] TestData = [] TestLabel = [] for i in range(int(TrainRate*m_jr)): Randindex = int(np.random.uniform(0, len(DataIndex))) TrainData.append(Data[:, Randindex]) TrainLabel.append(Label[:, Randindex]) del (DataIndex[Randindex]) for i in DataIndex: TestData.append(Data[:, i]) TestLabel.append(Label[:, i]) #对Urban随机采样 DataIndex = list(range(m_ur)) for i in range(int(TrainRate * m_ur)): Randindex = int(np.random.uniform(0, len(DataIndex))) TrainData.append(Data[:, Randindex + m_jr]) TrainLabel.append(Label[:, Randindex + m_jr]) del (DataIndex[Randindex]) for i in DataIndex: TestData.append(Data[:, i + m_jr]) TestLabel.append(Label[:, i + m_jr]) TrainLabel = np.array(TrainLabel) TestLabel = np.array(TestLabel) TrainData = np.array(TrainData) TestData = np.array(TestData) return TrainData, TrainLabel, TestData, TestLabel def combat_rnn_without_cc(objective, optimizer, metrics ): X = Input(shape=(157, 1)) lstm1 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.2))(X) lstm1 = BatchNormalization()(lstm1) lstm2 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.3))(lstm1) lstm2 = BatchNormalization()(lstm2) lstm3 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.4))(lstm2) lstm3 = BatchNormalization()(lstm3) X2 = Flatten()(lstm3) Global = Conv1D(3, 5, activation='relu', padding='same')(X) Global = MaxPooling1D(pool_size=2)(Global) Global = Conv1D(6, 4, activation='relu')(Global) Global = MaxPooling1D(pool_size=2)(Global) Global = Conv1D(12, 5, activation='relu')(Global) Global = MaxPooling1D(pool_size=2)(Global) Global = Conv1D(24, 4, activation='relu')(Global) Global = MaxPooling1D(pool_size=2)(Global) Global = Flatten()(Global) Con = concatenate([X2, Global]) Den1 = Dense(600, activation='relu', use_bias=None)(Con) Global = Dense(150, activation='relu', use_bias=None)(Den1) Abadunce = Dense(6, activation='softmax')(Global) model = Model(input=X, output=Abadunce) model.compile(loss=objective, optimizer=optimizer, metrics=metrics) return model def combat_rnn_with_cc(objective, optimizer, metrics ): X = Input(shape=(157, 1)) lstm1 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.2))(X) lstm1 = BatchNormalization()(lstm1) lstm2 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.3))(lstm1) lstm2 = BatchNormalization()(lstm2) lstm3 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.4))(lstm2) lstm3 = BatchNormalization()(lstm3) X2 = Flatten()(lstm3) Global = Conv1D(3, 5, activation='relu', padding='same')(X) Global = MaxPooling1D(pool_size=2)(Global) Global = Conv1D(6, 4, activation='relu')(Global) Global = MaxPooling1D(pool_size=2)(Global) Global = Conv1D(12, 5, activation='relu')(Global) Global = MaxPooling1D(pool_size=2)(Global) Global = Conv1D(24, 4, activation='relu')(Global) Global = MaxPooling1D(pool_size=2)(Global) Global = Flatten()(Global) Con = concatenate([X2, Global]) Den1 = Dense(600, activation='relu', use_bias=None)(Con) Global = Dense(150, activation='relu', use_bias=None)(Den1) Abadunce = Reshape((1, 150))(Global) Abadunce = Bidirectional(LSTM(100, dropout=0.2))(Abadunce) Abadunce = BatchNormalization()(Abadunce) Abadunce = RepeatVector(6)(Abadunce) Abadunce = Bidirectional(LSTM(50, return_sequences=True, kernel_regularizer=l2(0.0001), dropout=0.2))(Abadunce) Abadunce = BatchNormalization()(Abadunce) Abadunce = Bidirectional(LSTM(20, return_sequences=True, kernel_regularizer=l2(0.0001)))(Abadunce) Abadunce = TimeDistributed(Dense(1, activation='sigmoid'))(Abadunce) model = Model(input=X, output=Abadunce) model.compile(loss=objective, optimizer=optimizer, metrics=metrics) return model def pixel_rnn_without_cc(objective, optimizer, metrics): X = Input(shape=(157, 1)) lstm1 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.2))(X) lstm1 = BatchNormalization()(lstm1) lstm2 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.3))(lstm1) lstm2 = BatchNormalization()(lstm2) lstm3 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.4))(lstm2) lstm3 = BatchNormalization()(lstm3) X2 = Flatten()(lstm3) Den1 = Dense(600, activation='relu', use_bias=None)(X2) Global = Dense(150, activation='relu', use_bias=None)(Den1) Abadunce = Dense(6, activation='softmax')(Global) model = Model(input=X, output=Abadunce) model.compile(loss=objective, optimizer=optimizer, metrics=metrics) return model def pixel_rnn_with_cc(objective, optimizer, metrics): X = Input(shape=(157, 1)) lstm1 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.2))(X) lstm1 = BatchNormalization()(lstm1) lstm2 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.3))(lstm1) lstm2 = BatchNormalization()(lstm2) lstm3 = Bidirectional(LSTM(10, return_sequences=True, dropout=0.4))(lstm2) lstm3 = BatchNormalization()(lstm3) X2 = Flatten()(lstm3) Den1 = Dense(600, activation='relu', use_bias=None)(X2) Global = Dense(150, activation='relu', use_bias=None)(Den1) Abadunce = Reshape((1, 150))(Global) Abadunce = Bidirectional(LSTM(100, dropout=0.2))(Abadunce) Abadunce = BatchNormalization()(Abadunce) Abadunce =