Deep Architectures Fail to Generalize: A Lightweight Alternative for Agricultural Domain Transfer in Hyperspectral Images

Praveen Pankajakshan, Aravind Padmasanan, S. Sundar · 2025

Read Deep Architectures Fail to Generalize: A Lightweight Alternative for Agricultural Domain Transfer in Hyperspectral Images by Praveen Pankajakshan, Aravind Padmasanan, S. Sundar (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-12788353.txt.

Description

Praveen Pankajakshan, Aravind Padmasanan, and S. Sundar argue that giant neural networks are the wrong tool for labeling crops in satellite images from new regions, and offer a leaner one. Their 2025 paper builds a classifier for hyperspectral images that weighs spectral likeness and spatial nearness together. Farm scenes shift class balance, drop known crops, and add new ones between regions and seasons.

The method projects pixels onto a low-dimensional spectral manifold, drafts a probability map, then refines it with a kernel blending space and spectrum. It beat spectral-only models by 7 to 15 points and, with an unsupervised polish, passed several recent heavyweight systems. It needs only 1 to 10 percent labeled data and two tuning knobs. Transfers to Italian rice fields worked with no retraining.

Catalog record

CREATORPraveen Pankajakshan, Aravind Padmasanan, S. Sundar
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC12788353

Browse the full library · About the Tower of Babel