[1]:
from panel_segmentation import utils
import glob
import matplotlib.pyplot as plt
from PIL import Image
import os
import pandas as pd
import folium
import torch
from mmengine import Config
from mmdet.apis import init_detector, inference_detector
from mmdet.registry import VISUALIZERS
from shapely.geometry import Polygon
import mmcv
SOL-Searcher Pipeline#
The following example uses the SOL-Searcher pipeline to detect solar panels within a geographic bounding box. First, get the northwest and southeast latitude-longitude coordinates to generate a bounding box around an area. We can then grid the bounding box and pull a set of satellite images.
[2]:
# Pull an area of images in the Austin metro area to run through the SOL-Searcher model.
# Use the generate_satellite_imagery_grid() function to do this.
northwest_lat, northwest_lon = 30.304107, -97.708710
southeast_lat, southeast_lon = 30.299329, -97.697781
google_maps_api_key = "YOUR_API_KEY_HERE"
zoom_level = 19
file_save_folder = "../../../panel_segmentation/examples/sol-searcher_data_examples"
grid_info_list = utils.generateSatelliteImageryGrid(northwest_lat, northwest_lon,
southeast_lat, southeast_lon,
google_maps_api_key,
file_save_folder,
zoom_level,
lat_lon_distance=.0014)
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
File already pulled!
Visualize some of the imagery in the grid as reference.
[3]:
plot = utils.visualizeSatelliteImageryGrid(grid_info_list, file_save_folder)
plt.show()
30.304107_-97.70871.png
30.302707_-97.70871.png
30.301307_-97.70871.png
30.299907_-97.70871.png
30.298507_-97.70871.png
30.304107_-97.70731.png
30.302707_-97.70731.png
30.301307_-97.70731.png
30.299907_-97.70731.png
30.298507_-97.70731.png
30.304107_-97.70591.png
30.302707_-97.70591.png
30.301307_-97.70591.png
30.299907_-97.70591.png
30.298507_-97.70591.png
30.304107_-97.70451.png
30.302707_-97.70451.png
30.301307_-97.70451.png
30.299907_-97.70451.png
30.298507_-97.70451.png
30.304107_-97.70311.png
30.302707_-97.70311.png
30.301307_-97.70311.png
30.299907_-97.70311.png
30.298507_-97.70311.png
30.304107_-97.70171.png
30.302707_-97.70171.png
30.301307_-97.70171.png
30.299907_-97.70171.png
30.298507_-97.70171.png
30.304107_-97.70031.png
30.302707_-97.70031.png
30.301307_-97.70031.png
30.299907_-97.70031.png
30.298507_-97.70031.png
30.304107_-97.69891.png
30.302707_-97.69891.png
30.301307_-97.69891.png
30.299907_-97.69891.png
30.298507_-97.69891.png
30.304107_-97.69751.png
30.302707_-97.69751.png
30.301307_-97.69751.png
30.299907_-97.69751.png
30.298507_-97.69751.png
Let’s load in the SOL-searcher algorithm, run some imagery through it, and visualize the outputs.
[4]:
cfg = Config.fromfile("../../../panel_segmentation/models/sol_searcher_config.py")
checkpoint_file = "../../../panel_segmentation/models/sol_searcher_model.pth"
model = init_detector(cfg, device='cpu')
checkpoint = torch.load(checkpoint_file, map_location='cpu', weights_only=False)
model.load_state_dict(checkpoint['state_dict'])
visualizer = VISUALIZERS.build(model.cfg.visualizer)
# Visualize the first n images in example
n_images = 8
fig, axes = plt.subplots(((n_images + 1) // 2), 2, figsize=(10, 20))
axes = axes.ravel()
# Run inference on the example images
inference_data_list = list()
results = list()
grid_files = glob.glob(os.path.join(file_save_folder, "*.png"))
for i, file in enumerate(grid_files):
img = mmcv.imread(file, channel_order='rgb')
new_result = inference_detector(model, img)
new_result.file = file
results.append(new_result)
# Visualize n images
if i < n_images:
visualizer.add_datasample(
file,
img,
data_sample=new_result,
draw_gt=None,
wait_time=0,
pred_score_thr=0.35
)
img = visualizer.get_image()
axes[i].imshow(img)
axes[i].axis('off')
plt.tight_layout()
plt.show()
C:\Users\Quyen\Documents\all_venv\panel_seg_extreme_weather\Lib\site-packages\mmdet\apis\inference.py:70: UserWarning: checkpoint is None, use COCO classes by default.
warnings.warn('checkpoint is None, use COCO classes by default.')
C:\Users\Quyen\Documents\all_venv\panel_seg_extreme_weather\Lib\site-packages\mmdet\models\layers\se_layer.py:158: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
with torch.cuda.amp.autocast(enabled=False):
C:\Users\Quyen\Documents\all_venv\panel_seg_extreme_weather\Lib\site-packages\mmdet\models\backbones\csp_darknet.py:118: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
with torch.cuda.amp.autocast(enabled=False):
C:\Users\Quyen\Documents\all_venv\panel_seg_extreme_weather\Lib\site-packages\torch\functional.py:554: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\pytorch\aten\src\ATen\native\TensorShape.cpp:4316.)
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
Filter out the inference results with low prediction scores to reduce invalid detections and api queries.
[5]:
filtered_results = []
for result in results:
# Filter out inferences with prediction scores lower than 0.15
if hasattr(result, "pred_instances"):
save_inference = result.pred_instances.scores > 0.15
result.pred_instances = result.pred_instances[save_inference]
filtered_results.append(result)
Let’s visualize some of the result outputs from the model.
[6]:
# Loop through each inference result and get the lat-lon coordinates of the centroid of the box outputs
solar_location_list = list()
for result in filtered_results:
file_name = os.path.basename(result.file)
img_center_lat, img_center_lon = float(file_name.split("_")[0]), float(file_name.split("_")[-1].replace(".png", ""))
image_x_pixels, image_y_pixels = result.img_shape
boxes = result.pred_instances.bboxes
for box in boxes:
np_box = list(box.numpy())
box_lat, box_lon = utils.getInferenceBoxLatLonCoordinates(np_box, img_center_lat, img_center_lon,
image_x_pixels, image_y_pixels,
zoom_level)
solar_location_list.append((box_lat, box_lon))
If we want to associate individual installations to specific buildings, we can use Google’s geocoding API to get individual addresses for each PV installation.
[7]:
solar_install_address_list = list()
for lat_lon_coords in solar_location_list:
latitude, longitude = lat_lon_coords[0], lat_lon_coords[1]
# Look up the address associated with lat-lon coordinate
solar_install_address = utils.generateAddress(latitude, longitude, google_maps_api_key)
solar_install_address_list.append({"latitude": latitude,
"longitude": longitude,
"address": solar_install_address})
We can then take the median lat-lon coordinates by address as our main address-level lat-lon coordinates, and visualize the resulting dataframe.
[8]:
solar_address_df = pd.DataFrame(solar_install_address_list)
solar_address_df['median_latitude_address'] = solar_address_df.groupby("address")['latitude'].transform('median')
solar_address_df['median_longitude_address'] = solar_address_df.groupby("address")['longitude'].transform('median')
# Sort by address
solar_address_df = solar_address_df.sort_values(by='address')
solar_address_df.head(10)
[8]:
| latitude | longitude | address | median_latitude_address | median_longitude_address | |
|---|---|---|---|---|---|
| 195 | 30.304714 | -97.707270 | 1301 Barbara Jordan Blvd #200G, Austin, TX 787... | 30.304714 | -97.707270 |
| 200 | 30.304370 | -97.706632 | 1301 Barbara Jordan Blvd #200g, Austin, TX 787... | 30.304584 | -97.706632 |
| 194 | 30.304586 | -97.706803 | 1301 Barbara Jordan Blvd #200g, Austin, TX 787... | 30.304584 | -97.706632 |
| 193 | 30.304584 | -97.706620 | 1301 Barbara Jordan Blvd #200g, Austin, TX 787... | 30.304584 | -97.706632 |
| 201 | 30.304707 | -97.707157 | 1301 Barbara Jordan Blvd, Austin, TX 78723, USA | 30.304707 | -97.707157 |
| 180 | 30.302842 | -97.708257 | 1311 1/2 Philomena St, Austin, TX 78723, USA | 30.302842 | -97.708257 |
| 184 | 30.304479 | -97.703131 | 1500 Barbara Jordan Blvd, Austin, TX 78723, USA | 30.304479 | -97.703131 |
| 137 | 30.301051 | -97.705109 | 1501 1/2 Philomena St, Austin, TX 78723, USA | 30.301051 | -97.705109 |
| 182 | 30.303375 | -97.702519 | 1527 1/2 Barbara Jordan Blvd, Austin, TX 78723... | 30.303409 | -97.702622 |
| 188 | 30.303409 | -97.702622 | 1527 1/2 Barbara Jordan Blvd, Austin, TX 78723... | 30.303409 | -97.702622 |
Using aggregated address data, map the addresses containing solar installations in Folium. Please note, this map may include older satellite imagery and there may be discrepancies if solar installs are present in newer scans from Google Maps (i.e. solar was recently installed). Outputs of this algorithm should be manually reviewed to determine if there any discrepancies.
[9]:
solar_address_df_agg = solar_address_df[['address',
'median_latitude_address',
'median_longitude_address']].drop_duplicates()
# Create folium map to visualize the solar addresses
folium_map = folium.Map(location=[solar_address_df_agg['median_latitude_address'].mean(),
solar_address_df_agg['median_longitude_address'].mean()],
tiles = 'https://server.arcgisonline.com/ArcGIS/rest/services/World_Imagery/MapServer/tile/{z}/{y}/{x}',
attr = 'Esri',
name = 'Esri Satellite',
zoom_start=15,
max_zoom=19)
solar_feature_group = folium.FeatureGroup(name="solar", show=True)
for idx, row in solar_address_df_agg.iterrows():
lat = row['median_latitude_address']
lon = row['median_longitude_address']
folium.CircleMarker([lat, lon],
color='b',
weight=.05,
fill=True,
fill_opacity=0.6,
fill_color='red',
radius=3).add_to(solar_feature_group)
folium_map.add_child(solar_feature_group)
folium_map
[9]:
Make this Notebook Trusted to load map: File -> Trust Notebook
[ ]: