scripts/classify.py - master - DiHASTRO/pp-heat-map

scripts/classify.py
137 строк5 KB
Sm4rt1
Добавлена продвинутая классификация тепловых излучений
24 ноя 2025, 18:48
24 ноя 2025, 18:486479073
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans
from sklearn.neighbors import NearestNeighbors
import folium
import glob

files = glob.glob("scripts/static/*.csv")
dfs = [pd.read_csv(f) for f in files]
data = pd.concat(dfs, ignore_index=True)
print("Всего точек:", len(data))


data = data.dropna(subset=['bright_ti4', 'bright_ti5', 'frp'])
data['daynight'] = data['daynight'].map({'D': 1, 'N': 0})
data['confidence'] = data['confidence'].map({'h': 3, 'n': 2, 'l': 1}).fillna(0)


X = data[['bright_ti4', 'bright_ti5', 'frp', 'confidence', 'daynight']]
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

kmeans = KMeans(n_clusters=3, random_state=42)
data['cluster_intensity'] = kmeans.fit_predict(X_scaled)


coords = np.radians(data[['latitude', 'longitude']].values)
nbrs = NearestNeighbors(radius=5/6371.0, metric='haversine').fit(coords) 
distances, indices = nbrs.radius_neighbors(coords)
data['neighbor_count'] = [len(i) for i in indices]

data['std_local_frp'] = [np.std(data.iloc[idx]['frp']) if len(idx) > 1 else 0 for idx in indices]
data['std_local_ti4'] = [np.std(data.iloc[idx]['bright_ti4']) if len(idx) > 1 else 0 for idx in indices]
data['mean_local_frp'] = [np.mean(data.iloc[idx]['frp']) if len(idx) > 1 else data.iloc[i]['frp'] for i, idx in enumerate(indices)]

def classify_source(row):
    # --- 1. Промышленные факелы ---
    # стабильные, горячие источники с низкой изменчивостью FRP и Ti4
    if row['frp'] > 10 and row['std_local_ti4'] < 6 and row['std_local_frp'] < 8 and row['neighbor_count'] >= 2:
        return 'industrial_flare'

    # --- 2. Пожары ---
    # высокая FRP, большая изменчивость FRP и Ti4, группа близких точек
    elif row['frp'] > 10 and row['neighbor_count'] > 5 and (row['std_local_frp'] >= 8 or row['std_local_ti4'] >= 8):
        return 'fire_cluster'

    # --- 3. Городское тепло ---
    # слабое излучение, часто ночью, но с высокой плотностью точек
    elif row['frp'] < 10 and row['daynight'] == 0 and row['neighbor_count'] > 5:
        return 'heat'

    # --- 4. Слабые/единичные источники ---
    elif row['frp'] < 4 and row['neighbor_count'] <= 2:
        return 'weak_source'

    # --- 5. Всё остальное ---
    else:
        return 'unknown'


data['source_type'] = data.apply(classify_source, axis=1)


def smooth_label(i, idx_list):
    if len(idx_list) < 3:
        return data.loc[i, 'source_type']
    neighbor_types = data.iloc[idx_list]['source_type']
    main_type = neighbor_types.mode()[0]
    current_type = data.loc[i, 'source_type']
    if current_type != main_type and neighbor_types.value_counts().max() / len(neighbor_types) > 0.6:
        return main_type
    return current_type

data['source_type_smoothed'] = [
    smooth_label(i, idx) for i, idx in enumerate(indices)
]

changed = np.sum(data['source_type_smoothed'] != data['source_type'])
print(f"Исправлено точек: {changed} ({100*changed/len(data):.1f}%)")

data['source_type'] = data['source_type_smoothed']

def resolve_unknown(row):
    if row['source_type'] != 'unknown':
        return row['source_type']
    if row['frp'] > 35:
        return 'industrial_flare'
    elif row['frp'] > 10:
        return 'fire_cluster'
    elif row['daynight'] == 0 and row['neighbor_count'] > 2:
        return 'heat'
    elif row['frp'] < 3:
        return 'weak_source'
    else:
        return 'heat'
data['source_type'] = data.apply(resolve_unknown, axis=1)

print("\nТипы источников после коррекции:")
print(data['source_type'].value_counts())
print("\nСредние по типам:")
print(data.groupby('source_type')[['bright_ti4','frp','neighbor_count','std_local_frp']].mean())

m = folium.Map(location=[52.0, 52.9], zoom_start=6)


color_map = {
    'fire_cluster': 'red',
    'industrial_flare': 'orange',
    'heat': 'blue',
    'weak_source': 'gray',
}

for _, row in data.iterrows():
    radius = max(3, min(12, row['frp'] / 5))
    popup_html = f"""
    <b>Источник:</b> {row['source_type']}<br>
    <b>FRP:</b> {row['frp']:.1f} МВт<br>
    <b>bright_ti4:</b> {row['bright_ti4']:.1f} K<br>
    <b>bright_ti5:</b> {row['bright_ti5']:.1f} K<br>
    <b>Confidence:</b> {row['confidence']}<br>
    <b>Day/Night:</b> {'Day' if row['daynight']==1 else 'Night'}<br>
    <b>Neighbors (5 км):</b> {row['neighbor_count']}<br>
    <b>Std FRP (локально):</b> {row['std_local_frp']:.2f}<br>
    <b>Std Ti4 (локально):</b> {row['std_local_ti4']:.2f}<br>
    """

    folium.CircleMarker(
        [row['latitude'], row['longitude']],
        radius=radius,
        color=color_map.get(row['source_type'], 'black'),
        fill=True,
        fill_opacity=0.8,
        popup=folium.Popup(popup_html, max_width=300)
    ).add_to(m)

m.save("scripts/static/source_types_refined.html")