Sentinel2App / app.py
Dingyi6's picture
Update app.py
a28a432
import os
import sys
import time
import io
import json
from pathlib import Path
from utils import print_with_line_number
# Add the parent directory to the Python path
parent_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.append(parent_dir)
import ee
# Add Google Service account credential. Authenticates to the Earth Engine servers.
credentials = ee.ServiceAccountCredentials(os.environ.get("SERVICE_EMAIL"), key_data=os.environ.get("SERVICE_JSON"))
import math
from typing import Optional
from shiny import App, render, ui, reactive, Inputs, Outputs, Session, req
import ipyleaflet as L
from ipywidgets import Layout
from htmltools import css
import numpy as np
import pandas as pd
# from PIL import Image
from shinywidgets import output_widget, reactive_read, register_widget
from geopy.geocoders import Nominatim
import json
import requests
import traceback
from datetime import datetime, date
from typing import List
from timezonefinder import TimezoneFinder
# Library for ANN model loading
import tensorflow
import joblib
# Sentinel 2 Bands
# Sentinel-2 carries the Multispectral Imager (MSI). This sensor delivers 13 spectral bands ranging from 10 to 60-meter pixel size.
# Its blue (B2), green (B3), red (B4), and near-infrared (B8) channels have a 10-meter resolution.
# Next, its red edge (B5), near-infrared NIR (B6, B7, and B8A), and short-wave infrared SWIR (B11 and B12) have a ground sampling distance of 20 meters.
# Finally, its coastal aerosol (B1) and cirrus band (B10) have a 60-meter pixel size.
# Band Resolution Central Wavelength Description
# B1 60 m 443 nm Ultra Blue (Coastal and Aerosol)
# B2 10 m 490 nm Blue
# B3 10 m 560 nm Green
# B4 10 m 665 nm Red
# B5 20 m 705 nm Visible and Near Infrared (VNIR)
# B6 20 m 740 nm Visible and Near Infrared (VNIR)
# B7 20 m 783 nm Visible and Near Infrared (VNIR)
# B8 10 m 842 nm Visible and Near Infrared (VNIR)
# B8a 20 m 865 nm Visible and Near Infrared (VNIR)
# B9 60 m 940 nm Short Wave Infrared (SWIR)
# B10 60 m 1375 nm Short Wave Infrared (SWIR) - excluded
# B11 20 m 1610 nm Short Wave Infrared (SWIR)
# B12 20 m 2190 nm Short Wave Infrared (SWIR)
tf = TimezoneFinder()
# You can use different URLs to load remote sensing image data from various sources
# In this example, we use image data from Google Earth Engine
GEEurl = 'https://earthengine.googleapis.com/v1alpha/projects/earthengine-legacy/maps/{mapid}/tiles/{z}/{x}/{y}?token={token}'
GEEmap_id = '' # Replace with your Google Earth Engine Map ID
GEEtoken = '' # Replace with your Google Earth Engine Token
# Custom Loss Function with Covariance Penalty
def custom_loss(lam, cov_real_data):
def loss(y_true, y_pred):
mse_loss = tensorflow.reduce_mean(tensorflow.square(y_true - y_pred))
cov_pred = tensorflow.linalg.matmul(tensorflow.transpose(y_pred - tensorflow.reduce_mean(y_pred, axis=0)),
(y_pred - tensorflow.reduce_mean(y_pred, axis=0))) / tensorflow.cast(tensorflow.shape(y_pred)[0], tensorflow.float32)
cov_penalty = tensorflow.reduce_sum(tensorflow.square(cov_pred - cov_real_data))
return mse_loss + lam * cov_penalty
return loss
def load_model_and_preprocessors(model_path, cov_real_data_path, scaler_X_path, scaler_y_path):
# Load the covariance matrix
cov_real_data = np.load(cov_real_data_path)
# Load the trained model
model = tensorflow.keras.models.load_model(model_path, custom_objects={'loss': custom_loss(1e-6, cov_real_data)})
# Load the input data scaler
scaler_X = joblib.load(scaler_X_path)
# Load the output data scaler
scaler_Y = joblib.load(scaler_y_path)
return model, scaler_X, scaler_Y
# Load ANN model and preprocessors
model, loaded_scaler_X, loaded_scaler_Y = load_model_and_preprocessors(
"ANN_assests/model",
"ANN_assests/cov_real_data.npy",
"ANN_assests/scaler_X.pkl",
"ANN_assests/scaler_y.pkl"
)
# Create labels
labels = ['Longitude', 'Latitude', 'B1', 'B2', 'B3', 'B4', 'B5', 'B6', 'B7', 'B8', 'B8A', 'B9', 'B11', 'B12', 'N', 'Cab', 'Ccx', 'Cw', 'Cm']
X_labels = ['B1', 'B2', 'B3', 'B4', 'B5', 'B6', 'B7', 'B8', 'B8A', 'B9', 'B11', 'B12']
output_labels = ["N", "Cab", "Ccx", "Cw", "Cm"]
layer_names = ["structure parameter", "Chlorophylla+b content (µg/cm2)", "Carotenoids content (µg/cm2)", "Equivalent Water content (cm)", "Leaf Mass per Area (g/cm2)"]
data_to_map = {
"structure parameter": "N",
"Chlorophylla+b content (µg/cm2)": "Cab",
"Carotenoids content (µg/cm2)": "Ccx",
"Equivalent Water content (cm)": "Cw",
"Leaf Mass per Area (g/cm2)": "Cm"
}
# gradient_settings = {
# "structure parameter": {0.2: 'rgba(0, 0, 255, 1.0)', 0.6: 'rgba(0, 255, 255, 1.0)', 1.0: 'rgba(0, 255, 0, 1.0)'},
# "Chlorophylla+b content (µg/cm2)": {0.2: 'rgba(0, 128, 0, 1.0)', 0.6: 'rgba(127, 255, 0, 1.0)', 1.0: 'rgba(255, 255, 0, 1.0)'},
# "Carotenoids content (µg/cm2)": {0.2: 'rgba(255, 69, 0, 1.0)', 0.6: 'rgba(255, 0, 0, 1.0)', 1.0: 'rgba(139, 0, 0, 1.0)'},
# "Equivalent Water content (cm)": {0.2: 'rgba(0, 0, 139, 1.0)', 0.6: 'rgba(65, 105, 225, 1.0)', 1.0: 'rgba(0, 191, 255, 1.0)'},
# "Leaf Mass per Area (g/cm2)": {0.2: 'rgba(75, 0, 130, 1.0)', 0.6: 'rgba(148, 0, 211, 1.0)', 1.0: 'rgba(255, 20, 147, 1.0)'}
# }
gradient_parula = {
0.0: 'rgba(128, 0, 128, 1.0)', # purple
0.2: 'rgba(0, 0, 255, 1.0)', # blue
0.4: 'rgba(0, 255, 255, 1.0)', # cyan
0.5: 'rgba(0, 250, 154, 1.0)', # mediumspringgreen
0.6: 'rgba(50, 205, 50, 1.0)', # lime
0.7: 'rgba(173, 255, 47, 1.0)', # greenyellow
0.8: 'rgba(255, 255, 0, 1.0)', # yellow
0.9: 'rgba(255, 165, 0, 1.0)', # orange
1.0: 'rgba(255, 0, 0, 1.0)' # red
}
gradient_settings = {
"structure parameter": gradient_parula,
"Chlorophylla+b content (µg/cm2)": gradient_parula,
"Carotenoids content (µg/cm2)": gradient_parula,
"Equivalent Water content (cm)": gradient_parula,
"Leaf Mass per Area (g/cm2)": gradient_parula
}
print_with_line_number("Finish loading the ANN model!")
def runModel(input_data, scaler_X, scaler_Y, ANNmodel):
# Preprocess the Input Data
# Scale the input features using the previously saved scaler for X
input_data_scaled = scaler_X.transform(input_data)
# Use the Model for Prediction
# Predict the output values (N, Cab, Ccx, Cw, Cm) for each pixel block
output_data_scaled = ANNmodel.predict(input_data_scaled)
# Post-process the Output Data
# Inverse scale the output data using the previously saved scaler for Y
output_data = scaler_Y.inverse_transform(output_data_scaled)
# Organize the Output Results and Coordinates
# Create datasets for each output label and one for coordinates
# Each dataset contains corresponding data for all pixel blocks
datasets = {}
for i, label in enumerate(output_labels):
datasets[label] = output_data[:, i]
# Print the results for verification
for label, data in datasets.items():
print(label, data)
return datasets
gpsurl = 'https://www.googleapis.com/geolocation/v1/geolocate?key=' + os.environ.get("GMAP_TOKEN")
def getGPS():
GPSurl = gpsurl
data = {'homeMobileCountryCode': 310, 'homeMobileNetworkCode': 410, 'considerIp': 'True'}
response = requests.post(GPSurl, data=json.dumps(data))
result = json.loads(response.content)
return result
def get_location(lat, lon):
geolocator = Nominatim(timeout=120, user_agent="when-to-fly")
location = geolocator.reverse(f"{lat},{lon}")
return location.address
app_ui = ui.page_fluid(
ui.div(
ui.strong("Tips:"),
ui.br(),
ui.span("1.Click the polygon icon on the map to draw a polygon, the circular icon to mark a location, the line icon to measure distance, and the icon in the top right corner of the map to select the layers you want to display."),
ui.br(),
ui.span("2.After selecting an area, click the 'Analyze' button to analyze the leaf-level feature data for that area. The results are presented as heat maps, with brighter areas indicating values closer to the maximum."),
ui.br(),
ui.span("3.Currently, the analysis does not support multiple polygons. The application will only recognize the last polygoned area."),
ui.br(),
ui.span("4.After analyzing the data of the drawn area, the webpage may experience slower loading speeds and delays. Please be patient and wait after performing an operation."),
ui.br(),
ui.strong("If you are unable to zoom in or out of the map using the mouse scroll wheel, please use the slide bar provided above to zoom directly.", style="color: green;"),
ui.br(),
ui.strong("We strongly recommend that you use a smaller scale to view the heat map (17 or 18 zoom level), as it will retain more details.", style="color: red;"),
ui.br(),
ui.strong("Please do not use the computer's touchscreen to zoom in on the map, as this can cause errors.", style="color: red;")
),
ui.layout_sidebar(
ui.panel_sidebar(
ui.div(
ui.div(
ui.input_date("date", "Date:"),
ui.input_slider("zoom", "Map zoom level", value=12, min=1, max=18),
),
ui.div(
ui.input_numeric("lat", "Latitude", value=38.53667742),
ui.input_numeric("long", "Longitude", value=-121.75387309),
),
style=css(display="flex", justify_content="center", align_items="center", gap="1rem"),
),
),
ui.panel_main(
ui.div(
ui.output_text("N_range"),
ui.output_text("Cab_range"),
ui.output_text("Ccx_range"),
ui.output_text("Cw_range"),
ui.output_text("Cm_range"),
style=css(display="flex", justify_content="center", align_items="center", gap="2rem"),
),
ui.img(src="legend.png"),
),
),
output_widget("map"),
ui.strong("Must analyze (to renew the image information) before downloading any file.", style="color: green;"),
ui.div(
ui.input_action_button("analyze", "Analyze", class_="btn-success"),
ui.download_button("download_polygon", "Download spectral data as tif", class_="btn-success"),
ui.download_button("download_output", "Download spectral and output data as csv", class_="btn-success"),
style=css(display="flex", justify_content="center", align_items="center", gap="2rem"),
),
)
# re-run when a user using the application
def server(input, output, session):
# Initialize Earth Engine
ee.Initialize(credentials)
global address_line, polygoned_image, output_df
address_line = None
polygoned_image = None
polygon_data = reactive.Value([])
output_df = pd.DataFrame()
N = reactive.Value("structure parameter")
Cab = reactive.Value("Chlorophylla+b content (µg/cm2)")
Ccx = reactive.Value("Carotenoids content (µg/cm2)")
Cw = reactive.Value("Equivalent Water content (cm)")
Cm = reactive.Value("Leaf Mass per Area (g/cm2)")
m = ui.modal(
"Please wait for progress...",
easy_close=False,
size="s",
footer=None,
fade=True
)
@output
@render.text
def N_range():
return N.get()
@output
@render.text
def Cab_range():
return Cab.get()
@output
@render.text
def Ccx_range():
return Ccx.get()
@output
@render.text
def Cw_range():
return Cw.get()
@output
@render.text
def Cm_range():
return Cm.get()
def handle_draw(self, action, geo_json):
print("运行handle_draw")
if geo_json['type'] == 'Feature':
# Check if the drawn shape is a polygon
if geo_json['geometry']['type'] == 'Polygon':
# Get the coordinates of the polygon's vertices
coordinates = geo_json['geometry']['coordinates'][0]
# Extract latitude and longitude values from each vertex
# For GeoJSON, coordinates are represented as [longitude, latitude]
# (note the reverse order compared to traditional [latitude, longitude])
polygon_data.set([(lon, lat) for lon, lat in coordinates])
# Process the polygon_data as per your requirement
# For example, print the coordinates
print("Polygon Vertex Coordinates:")
for lon, lat in polygon_data.get():
print(f"Latitude: {lat}, Longitude: {lon}")
ui.modal_show(m)
# Check API status
asset_roots = ee.data.getAssetRoots()
if asset_roots:
print("API is connected and working.")
else:
print("API is not connected or not working.")
try:
# Get the user's current geoinformation
current_gps = getGPS()
print_with_line_number(current_gps)
current_location = get_location(current_gps['location']['lat'], current_gps['location']['lng'])
print_with_line_number(current_location)
ui.update_text(id="address",
label="Data for",
value=current_location)
# Initialize and display when the session starts (1)
map = L.Map(center=(current_gps['location']['lat'], current_gps['location']['lng']), zoom=12, scroll_wheel_zoom=True)
map.layout = Layout(height='600px')
@reactive.isolate()
def update_text_inputs(lat: Optional[float], long: Optional[float]) -> None:
req(lat is not None, long is not None)
lat = round(lat, 8)
long = round(long, 8)
if lat != input.lat():
input.lat.freeze()
ui.update_text("lat", value=lat)
if long != input.long():
input.long.freeze()
ui.update_text("long", value=long)
map.add_layer(L.TileLayer(url='https://mt1.google.com/vt/lyrs=s&x={x}&y={y}&z={z}', name='Natural Map'))
# Add a distance scale
map.add_control(L.leaflet.ScaleControl(position="bottomleft"))
layer_control = L.LayersControl(position='topright')
map.add_control(layer_control)
# Add the DrawControl widget to the map
draw_control = L.DrawControl(
polygon = {
"shapeOptions": {
"fillColor": "transparent",
"fillOpacity": 0.0
}
}
)
map.add_control(draw_control)
# Attach the handle_draw function to the on_draw event
draw_control.on_draw(handle_draw)
register_widget("map", map)
ui.modal_remove()
except Exception as e:
ui.modal_remove()
error_modal = ui.modal(
str(e),
title="An Error occured. Please refresh",
easy_close=True,
size="xl",
footer=None,
fade=True
)
# print_with_line_number("Show error modal")
ui.modal_show(error_modal)
traceback.print_exc()
# When the slider changes, update the map's zoom attribute (2)
@reactive.Effect
def _():
map.zoom = input.zoom()
# When zooming directly on the map, update the slider's value (2 and 3)
@reactive.Effect
def _():
ui.update_slider("zoom", value=reactive_read(map, "zoom"))
@reactive.Effect
def location():
"""Returns tuple of (lat,long) floats--or throws silent error if no lat/long is
selected"""
# Require lat/long to be populated before we can proceed
req(input.lat() is not None, input.long() is not None)
try:
long = input.long()
# Wrap longitudes so they're within [-180, 180]
long = (long + 180) % 360 - 180
if round(map.center[0], 8) == input.lat() and round(map.center[1], 8) == long:
return
map.center = (input.lat(), long)
except ValueError as e:
error_modal = ui.modal(
str(e),
title="Invalid latitude/longitude specification. Please refresh",
easy_close=True,
size="xl",
footer=None,
fade=True
)
# print_with_line_number("Show error modal")
ui.modal_show(error_modal)
traceback.print_exc()
# Everytime the map's bounds change, update the output message (3)
# rerun when a user do some reactive changes.
@reactive.Effect
def map_bounds():
center = reactive_read(map, "center")
if len(center) == 0:
return
lon = (center[1] + 180) % 360 - 180
update_text_inputs(center[0], lon)
def update_or_create_heatmaps(output_datasets, scale):
"""
Check if a heatmap layer exists for each dataset in output_datasets.
If it exists, update the heatmap, otherwise create a new heatmap.
Parameters:
output_datasets (list of dict): The datasets for creating/updating heatmaps
"""
# Iterate over each dataset in output_datasets
existing_layers = {layer.name: layer for layer in map.layers}
print_with_line_number(existing_layers)
for layer_name in layer_names:
# Check if a heatmap layer with this name already exists
if layer_name in existing_layers:
print("deleting ", layer_name)
map.remove_layer(existing_layers[layer_name])
heatmap_data = []
data_values = output_datasets[data_to_map[layer_name]]
q03 = np.percentile(data_values, 3)
q97 = np.percentile(data_values, 97)
min_value = min(data_values)
if min_value < 0:
min_value = 0
max_value = max(data_values)
if (data_to_map[layer_name] == "N"):
N.set(layer_name + ": " + str(min_value) + " ~ " + str(max_value))
elif (data_to_map[layer_name] == "Cab"):
Cab.set(layer_name + ": " + str(min_value) + " ~ " + str(max_value))
elif (data_to_map[layer_name] == "Ccx"):
Ccx.set(layer_name + ": " + str(min_value) + " ~ " + str(max_value))
elif (data_to_map[layer_name] == "Cw"):
Cw.set(layer_name + ": " + str(min_value) + " ~ " + str(max_value))
else:
Cm.set(layer_name + ": " + str(min_value) + " ~ " + str(max_value))
for coord, n in zip(output_datasets["Coordinates"], data_values):
# normalized_value = (n - min_value) / (max_value - min_value)
if n <= q03:
normalized_value = 0
elif n >= q97:
normalized_value = 1
else:
normalized_value = (n - q03) / (q97 - q03)
heatmap_data.append([coord[1], coord[0], normalized_value])
# Generate new heatmap for this dataset
heatmap = L.Heatmap(
locations=heatmap_data,
radius=scale * 1.4,
gradient=gradient_settings[layer_name],
max=1,
blur=scale / 2,
name=layer_name
)
# Add the new heatmap layer to the map
map.add_layer(heatmap)
@reactive.Effect
@reactive.event(input.analyze, ignore_none=True, ignore_init=True)
def _():
global output_df
if not polygon_data.get():
return
ui.modal_show(m)
global polygoned_image
polygon = ee.Geometry.Polygon(polygon_data.get())
print("Polygon Data: " , polygon_data.get())
print("Polygon: " , polygon)
# Define Sentinel-2 image collection ("2021-01-01", "2021-12-31")
current_date = input.date()
today = ee.Date(input.date().strftime('%Y-%m-%d'))
start_date = today.advance(-15, 'day')
print("Start Date: ", start_date.format('YYYY-MM-dd').getInfo(), "| End Date: ", today.format('YYYY-MM-dd').getInfo())
sentinel2 = ee.ImageCollection("COPERNICUS/S2_SR")\
.filterDate(start_date, today)\
.filterBounds(polygon)\
.sort('CLOUDY_PIXEL_PERCENTAGE', True)
# .first() # Retrieve the first image from the ImageCollection
# sentinel2 = sentinel2.sort('CLOUDY_PIXEL_PERCENTAGE', True)
polygoned_image = sentinel2.first()
# Make sure the polygoned_image is available
if not polygoned_image:
print("No image available for download.")
return
retry = 5
while(sentinel2.size().getInfo() == 0):
if(retry == 0):
ui.update_date("date", label="Date:", value=current_date)
ui.modal_remove()
error_modal = ui.modal(
"Please do not choose a date that is too far in the future. This application will search for remote sensing data within the two weeks prior to the selected date.",
title="Something wrong happened, try \"Analyze\" again ",
easy_close=True,
size="xl",
footer=None,
fade=True
)
ui.modal_show(error_modal)
print("fail to fecth image.")
return
print("wait for fetching.")
time.sleep(2)
retry -= 1
print_with_line_number("Type of sentinel2: " + str(type(sentinel2)))
print("Counts of Fetched image: ", sentinel2.size().getInfo())
# Clip the image to the extent of the polygon
clipped_image = polygoned_image.clip(polygon)
# Get meta data about the image object
# bands = clipped_image.bandNames().getInfo()
# print_with_line_number(bands)
# Calculate suitable pixel number. GEE service allow fecthing 5000 pixels at most for one call. So we use the "polygoned area / 4999" to decide a rational pixel scale.
scale=1
polygon_area = polygon.area().getInfo()
num = math.ceil(polygon_area / scale / scale)
if (num > 4999):
per_area = math.ceil(polygon_area / 4998)
scale = math.ceil(math.pow(per_area, 1.0/2))
print("polygon_area(m2): ", polygon_area, "scale: ", scale)
# Fetch reflectance of B1-B12
spectral_values = clipped_image.select('B1', 'B2', 'B3', 'B4', 'B5', 'B6', 'B7', 'B8', 'B8A', 'B9', 'B11', 'B12').sample(
region=polygon,
scale=scale,
numPixels=4999,
geometries=True
)
print_with_line_number("Pre-process the bands data.")
# print_with_line_number(type(spectral_values))
spectral_values = spectral_values.getInfo()
# print_with_line_number(type(spectral_values))
spectral_values_json = json.dumps(spectral_values)
# print(spectral_values_json)
spectral_values_dict = json.loads(spectral_values_json)
features = spectral_values_dict['features']
print_with_line_number("Extract the center coordinates and values of B1-B12 for each pixel block")
coords = []
input_data = []
for feature in features:
coords.append(feature['geometry']['coordinates'])
props = feature['properties']
input_data.append([props[b] for b in X_labels])
# Convert to NumPy arrays
coords = np.array(coords)
# print("coords: ", coords)
input_data = np.array(input_data)
# print("input_bands: ", input_data)
output_datasets = runModel(input_data, loaded_scaler_X, loaded_scaler_Y, model)
# Combine all data
data_combined = np.column_stack((coords, input_data, output_datasets['N'], output_datasets['Cab'], output_datasets['Ccx'], output_datasets['Cw'], output_datasets['Cm']))
# Convert to DataFrame
output_df = pd.DataFrame(data_combined, columns=labels)
print_with_line_number("Add a dataset for the coordinates")
output_datasets['Coordinates'] = coords
update_or_create_heatmaps(output_datasets, scale)
register_widget("map", map)
ui.modal_remove()
@reactive.Effect
def _():
print("Current navbar page: ", input.navbar_id())
@session.download(
filename=lambda: f"image-{input.date().isoformat()}-{np.random.randint(100, 999)}.tif"
)
async def download_polygon():
# # Replace this with your ee.Image object
# image_id = "COPERNICUS/S2_SR/20230728T184921_20230728T190044_T10SFH"
# image = ee.Image(image_id)
# Make sure the polygoned_image is available
if not polygoned_image:
print("No image available for download.")
return
# Clip the image to the extent of the polygon
clipped_image = polygoned_image.clip(ee.Geometry.Polygon(polygon_data.get()))
print("clipped_image: ", clipped_image)
# Define export parameters
download_params = {
'scale': 10,
'region': polygon_data.get(), # ee.Geometry object defining the region to export
'format': 'GeoTIFF',
}
# Generate download URL for the GeoTIFF image
download_url = clipped_image.getDownloadURL(download_params)
# Send a request to download the image
response = requests.get(download_url)
# Create a BytesIO buffer
with io.BytesIO() as buf:
# Write the image content to the buffer
buf.write(response.content)
buf.seek(0) # Move the buffer's position to the beginning
# Yield the buffer's content as a downloadable file
yield buf.getvalue()
print("Image downloaded successfully!")
@session.download(
filename=lambda: f"data-{input.date().isoformat()}-{np.random.randint(100, 999)}.csv"
)
def download_output():
global output_df
# Check if data is available
if output_df.empty:
print("No data available for download.")
return
# Convert dataframe to CSV and encode to bytes
csv_data = output_df.to_csv(index=False).encode()
# Create a StringIO buffer for textual data
with io.BytesIO() as buf:
buf.write(csv_data)
# Reset the buffer's position to the beginning
buf.seek(0)
# Create and return a streaming response
yield buf.getvalue()
print("Data downloaded successfully!")
static_dir = Path(__file__).parent / "assets"
app = App(app_ui, server, static_assets=static_dir)