Deep Learning for Remote Sensing

import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report
# Pixel classification with spectral features (synthetic Sentinel-2 like bands)
rng=np.random.default_rng(7)
def sample(n,mu): return rng.normal(mu,0.03,(n,4))   # blue, green, red, nir
X=np.vstack([sample(300,[.05,.06,.04,.02]),sample(300,[.04,.08,.05,.40]),sample(300,[.12,.13,.15,.20])]); y=np.repeat([0,1,2],300)
ndvi=((X[:,3]-X[:,2])/(X[:,3]+X[:,2]))[:,None]; X=np.hstack([X,ndvi])
idx=rng.permutation(900); tr,te=idx[:600],idx[600:]
m=RandomForestClassifier(200,random_state=0).fit(X[tr],y[tr])
print(classification_report(y[te],m.predict(X[te]),target_names=["water","vegetation","urban"]))