遥感landsat 影像地物分类，基于CNN深度学习的方法.zip

from glob import glob import os, math, random, glob, time random.seed(2) import numpy as np import tensorflow as tf from sklearn.utils import resample from tensorflow.keras import layers from tensorflow import keras from tensorflow.keras.layers import Dense, Conv2D, MaxPooling2D, Dropout, Flatten from sklearn.metrics import confusion_matrix, precision_score, recall_score, f1_score #import matplotlib.pyplot as plt ######################################################################### ##### PART - B: READING NUMPY ARRAYS. TRAINING AND SAVING THE MODEL ##### ######################################################################### os.chdir(r"F:\PycharmProjects\img_class\Remote-sensing-CNN-land-classify") out_model='7class' # Load arrays from .npy files features = np.load('CNN_3by3_features.npy') labels = np.load('CNN_3by3_labels.npy') # Separate and balance the classes # forest forest_features = features[labels==0] forest_labels = labels[labels==0] # water water_features = features[labels==1] water_labels = labels[labels==1] # gress gress_features = features[labels==2] gress_labels = labels[labels==2] # ground ground_features = features[labels==3] ground_labels = labels[labels==3] built_features = features[labels==4] built_labels = labels[labels==4] # road road_features = features[labels==5] road_labels = labels[labels==5] out_features = features[labels==6] out_labels = labels[labels==6] print('Number of records in each class:') print('forest: %d, water: %d, gress:%d' % (forest_features.shape[0], water_features.shape[0],gress_features.shape[0])) # Downsample the majority class forest_features = resample(forest_features, replace = False, # sample without replacement n_samples = gress_features.shape[0], # match minority n random_state = 2) forest_labels = resample(forest_labels, replace = False, # sample without replacement n_samples = gress_labels.shape[0], # match minority n random_state = 2) ground_features = resample(ground_features, replace = False, # sample without replacement n_samples = gress_features.shape[0], # match minority n random_state = 2) ground_labels = resample(ground_labels, replace = False, # sample without replacement n_samples = gress_labels.shape[0], # match minority n random_state = 2) # print('Number of records in balanced classes:') # print('Built: %d, Unbuilt: %d' % (built_labels.shape[0], water_labels.shape[0])) print('resample') print('forest: %d, water: %d, gress:%d, groud:%d, build:%d, road:%d' % (forest_features.shape[0], water_features.shape[0],gress_features.shape[0],ground_features.shape[0],built_features.shape[0],road_features.shape[0])) # Combine the balanced features features = np.concatenate((forest_features, water_features,gress_features,ground_features,built_features,road_features,out_features), axis=0) labels = np.concatenate((forest_labels, water_labels,gress_labels,ground_labels,built_labels,road_labels,out_labels), axis=0) forest_sum=forest_features.shape[0] water_sum=water_features.shape[0] gress_sum=gress_features.shape[0] ground_sum=ground_features.shape[0] build_sum=built_features.shape[0] road_sum=road_features.shape[0] out_sum=out_features.shape[0] feature_pos=[forest_sum,water_sum,gress_sum,ground_sum,build_sum,road_sum,out_sum] # Normalise the features 图片最大值为2047，应当除以2047进行归一化，以图片最大值，不是特征最大值 print(features.min(),features.max()) features = features*4 / (255.0*7) print('New values in input features, min: %d & max: %d' % (features.min(), features.max())) def feature_clip(featurepos,trainProp,order=1,labels=None): """ average clip each feature :param featurepos: input different feature position :param trainProp: rate of trainging dataset :param order: default is training dataset, otherwise is test dataset :return: """ clip=np.array([]) i=0 if order==1: for item in featurepos: index=np.arange(i,int((item)*trainProp)+i) clip=np.concatenate((clip,index),axis=0) i+=item else: for item in featurepos: index=np.arange(int((item)*trainProp)+i,item+i) clip=np.concatenate((clip,index),axis=0) i += item return clip # Define the function to split features and labels def train_test_split(features, labels, trainProp=0.7): dataSize = features.shape[0] sliceIndex = int(dataSize*trainProp) randIndex = np.arange(dataSize) # random.shuffle(randIndex) # a0=randIndex[:sliceIndex] # a=[randIndex[:sliceIndex]] trainx = features[[randIndex[:sliceIndex]], :, :, :][0] train=feature_clip(feature_pos,trainProp,1) test=feature_clip(feature_pos,trainProp,0) train_x=features[[train.astype(np.int32)], :, :, :][0] test_x = features[[test.astype(np.int32)], :, :, :][0] train_y = labels[train.astype(np.int32)] test_y = labels[test.astype(np.int32)] return(train_x, train_y, test_x, test_y) # Call the function to split the data train_x, train_y, test_x, test_y = train_test_split(features, labels) print('train data shape') print(train_x.shape[0]) print('test data shape') print(test_x.shape[0]) print(train_x.max()) print('Reshaped features:', train_x.shape, test_x.shape) _, rowSize, colSize, nBands = train_x.shape # Create a model model = keras.Sequential() model.add(Conv2D(32,strides=1, kernel_size=1, padding='valid', activation='relu', input_shape=(rowSize, colSize, nBands))) model.add(Dropout(0.25)) model.add(Conv2D(64,strides=1, kernel_size=1, padding='valid', activation='relu')) model.add(Dropout(0.25)) model.add(Conv2D(64,strides=1, kernel_size=1, padding='valid', activation='relu')) model.add(Dropout(0.25)) # model.add(MaxPooling2D((2,2),strides=2)) model.add(Flatten()) model.add(Dense(256, activation='relu')) model.add(Dense(128, activation='relu')) # model.add(Dropout(0.5)) model.add(Dense(7, activation='softmax')) model.compile(loss='sparse_categorical_crossentropy', optimizer= keras.optimizers.RMSprop(learning_rate=0.001),metrics=['accuracy']) history = model.fit(train_x, train_y, epochs=20,verbose=1,steps_per_epoch=2) # model = tf.keras.models.load_model('F:\PycharmProjects\img_class\Landsat-Classification-Using-Convolution-Neural-Network-master\cnn' # '\\trained_models\\CNN_7class_3by3.h5') print(model.summary()) # Predict for test data yTestPredicted = model.predict(test_x,steps=1) # yTestPredicted = yTestPredicted[:,-1] classnum=np.argmax(yTestPredicted,axis=1) model.save('CNN_%s_3by3.h5' % (out_model)) # Calculate and display the error metrics # yTestPredicted = (yTestPredicted>0.5).astype(int) cMatrix = confusion_matrix(test_y,classnum) # cMatrix = confusion_matrix(test_y, yTestPredicted) print("Confusion matrix:\n", cMatrix) # pScore = precision_score(test_y, yTestPredicted,average=None) pScore = precision_score(test_y, classnum,average=None) rScore = recall_score(test_y, classnum,average=None) fScore = f1_score(test_y, classnum,average=None)