Text this: Multiclass Crop Interpretation via a Lightweight Attentive Feature Fusion Network Using Vehicle-View Images