基于Attention U-Net与SAR影像的滑坡识别

主要参考这篇文献：Nava, L.; Bhuyan, K.; Meena, S.R.; Monserrat, O.; Catani, F. Rapid Mapping of Landslides on SAR Data by Attention U-Net. Remote Sens. 2022, 14, 1449.
滑坡是全球范围内常见的自然灾害，对区域发展与人类活动造成较大威胁。传统的滑坡监测和制图方法主要依赖于光学地球观测（EO）数据，但在云层遮蔽、夜间或极端天气条件下，光学影像的获取和使用受到严重限制。案例通过使用合成孔径雷达（SAR）数据，通过SAR能够在任何天气条件下和夜间获取地表信息的特点，并结合深度学习技术，实现在恶劣天气条件下快速识别滑坡的可能性。
这里采用的模型为Attention U-Net（Attn-U-Net），是一种改进的深度学习模型，基于经典的U-Net架构，通过引入注意力机制（Attention Mechanism），增强了模型对重要特征的关注能力，从而提高了分割的准确性和鲁棒性。
资料来源

安装依赖

• segmentation-models是一个基于PyTorch的图像分割库，提供了多种强大的模型架构（如U-Net、DeepLabV3+等），用于精确识别图像中的每个像素属于哪个目标。
• rasterio是一个用于处理栅格数据的Python库，基于GDAL。
• pencv-python是OpenCV的Python接口，是一个开源的计算机视觉库，提供了丰富的图像和视频处理功能。

!pip install segmentation-models rasterio -i https://mirrors.cloud.tencent.com/pypi/simple
!pip install opencv-python -i https://mirrors.cloud.tencent.com/pypi/simple

模型构建

构建模型，包括卷积层（Conv2D）、池化层（MaxPooling2D）、上采样层（UpSampling2D）、批归一化层（BatchNormalization）、Dropout层等

import cv2 
import time
import os
import h5py
import pandas as pd
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, concatenate, Conv2D, MaxPooling2D, Conv2DTranspose
from tensorflow.keras.layers import Activation, add, multiply, Lambda
from tensorflow.keras.layers import AveragePooling2D, average, UpSampling2D, Dropout
from tensorflow.keras.optimizers import Adam, SGD, RMSprop
from tensorflow.keras.initializers import glorot_normal, random_normal, random_uniform
from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard, EarlyStopping

from tensorflow.keras import backend as K
from tensorflow.keras.layers import BatchNormalization 
from tensorflow.keras.applications import VGG19, densenet
from tensorflow.keras.models import load_model

import tensorflow as tf
import numpy as np
from matplotlib import pyplot as plt
import segmentation_models as sm
from tensorflow.keras.metrics import MeanIoU
import random
import rasterio
from rasterio.plot import show
from sklearn.model_selection import train_test_split
from tensorflow.keras import layers
from tensorflow import keras
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, UpSampling2D, concatenate, Conv2DTranspose, BatchNormalization, Dropout, Lambda
from tensorflow.keras import backend as K

K.set_image_data_format('channels_last')  
kinit = 'glorot_normal'


def expend_as(tensor, rep,name):
	my_repeat = Lambda(lambda x, repnum: K.repeat_elements(x, repnum, axis=3), arguments={
   
   'repnum': rep},  name='psi_up'+name)(tensor)
	return my_repeat

# 定义注意力门控块函数
def AttnGatingBlock(x, g, inter_shape, name):
    shape_x = K.int_shape(x)  # 32
    shape_g = K.int_shape(g)  # 16

    theta_x = Conv2D(inter_shape, (2, 2), strides=(2, 2), padding='same', name='xl'+name)(x)  # 16
    shape_theta_x = K.int_shape(theta_x)

    phi_g = Conv2D(inter_shape, (1, 1), padding='same')(g)
    upsample_g = Conv2DTranspose(inter_shape, (3, 3),strides=(shape_theta_x[1] // shape_g[1], shape_theta_x[2] // shape_g[2]),padding='same', name='g_up'+name)(phi_g)  # 16

    concat_xg = add([upsample_g, theta_x])
    act_xg = Activation('relu')(concat_xg)
    psi = Conv2D(1, (1, 1), padding='same', name='psi'+name)(act_xg)
    sigmoid_xg = Activation('sigmoid')(psi)
    shape_sigmoid = K.int_shape(sigmoid_xg)
    upsample_psi = UpSampling2D(size=(shape_x[1] // shape_sigmoid[1], shape_x[2] // shape_sigmoid[2]))(sigmoid_xg)  # 32

    upsample_psi = expend_as(upsample_psi, shape_x[3],  name)
    y = multiply([upsample_psi, x], name='q_attn'+name)

    result = Conv2D(shape_x[3], (1, 1), padding='same',name='q_attn_conv'+name)(y)
    result_bn = BatchNormalization(name='q_attn_bn'+name)(result)
    return result_bn

# 定义U-Net中的卷积块函数
def UnetConv2D(input, outdim, is_batchnorm, name):
	x = Conv2D(outdim, (3, 3), strides=(1, 1), kernel_initializer=kinit