Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agricultural Landscape Understanding At Country-Scale

Radhika Dua, Nikita Saxena, Aditi Agarwal, Alex Wilson, Gaurav Singh, Hoang Tran, Ishan Deshpande, Amandeep Kaur, Gaurav Aggarwal, Chandan Nath, Arnab Basu, Vishal Batchu, Sharath Holla, Bindiya Kurle, Olana Missura, Rahul Aggarwal, Shubhika Garg, Nishi Shah, Avneet Singh, Dinesh Tewari, Agata Dondzik, Bharat Adsul, Milind Sohoni, Asim Rama Praveen, Aaryan Dangi, Lisan Kadivar, E Abhishek, Niranjan Sudhansu, Kamlakar Hattekar, Sameer Datar, Musty Krishna Chaithanya, Anumas Ranjith Reddy, Aashish Kumar, Betala Laxmi Tirumala, Alok Talekar

Abstract

Agricultural landscapes are quite complex, especially in the Global South where fields are smaller, and agricultural practices are more varied. In this paper we report on our progress in digitizing the agricultural landscape (natural and man-made) in our study region of India. We use high resolution imagery and a UNet style segmentation model to generate the first of its kind national-scale multi-class panoptic segmentation output. Through this work we have been able to identify individual fields across 151.7M hectares, and delineating key features such as water resources and vegetation. We share how this output was validated by our team and externally by downstream users, including some sample use cases that can lead to targeted data driven decision making. We believe this dataset will contribute towards digitizing agriculture by generating the foundational baselayer.

Agricultural Landscape Understanding At Country-Scale

Abstract

Agricultural landscapes are quite complex, especially in the Global South where fields are smaller, and agricultural practices are more varied. In this paper we report on our progress in digitizing the agricultural landscape (natural and man-made) in our study region of India. We use high resolution imagery and a UNet style segmentation model to generate the first of its kind national-scale multi-class panoptic segmentation output. Through this work we have been able to identify individual fields across 151.7M hectares, and delineating key features such as water resources and vegetation. We share how this output was validated by our team and externally by downstream users, including some sample use cases that can lead to targeted data driven decision making. We believe this dataset will contribute towards digitizing agriculture by generating the foundational baselayer.

Paper Structure

This paper contains 35 sections, 15 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Core Contributions
  3. Related Work
  4. Agricultural Landscape Understanding
  5. Overview
  6. Dataset Creation
  7. Labeling Process
  8. Dataset Preparation
  9. Task Description
  10. Classes and Layers
  11. Machine Learning model
  12. Model Architecture
  13. Training Objective
  14. Model Selection Criteria
  15. Post Processing
  16. ...and 20 more sections

Figures (15)

  • Figure 1: Demonstration of the model's ability to extract diverse agricultural features from high-resolution satellite imagery. (Left) Model input: High-resolution satellite imagery. (Right) Model output: Segmented instances of fields, trees, water bodies.
  • Figure 2: This figure illustrates the workflow of an ALU system designed to handle the challenges of national-scale agricultural land use mapping. (a) Complete coverage requires multiple satellite images acquired at different times ($t_{1}$ to $t_{n}$) due to the large study area and satellite orbital constraints. (b) To ensure scalability, the processing is spatially partitioned using S2 cells, a gridded sub-region of the Earth. (c-e) A machine learning model performs multi-class panoptic segmentation on each image to identify agricultural features. (f-h) Post-processing steps, including vectorization, de-duplication, and merging, reconcile information from overlapping images and generate a final output partitioned at s2 cell
  • Figure 3: Annotators were provided guidelines to segment the landscape within the red box (i.e., the tile on the left) into the given categories (right).
  • Figure 4: Example of an annotated sample. Left: input high resolution satellite image. Right: human labeled annotations
  • Figure 5: Visualizing the Task: This figure illustrates our objective: given an input satellite image with RGB bands (left), our goal is to generate both a semantic segmentation map and an instance segmentation map. The center and right images showcase these outputs for the ground instance layer, highlighting the distinction between class-level and object-level segmentation.
  • ...and 10 more figures