Airborne monitoring system for in-season agriculture : operational considerations and image processing for wide-area sensing

Abstract

Remote sensing, in particular multispectral imagery, can measure crop health and detect in-season disturbances such as pests and diseases before they are visible to the naked eye, but it is inaccessible to small-plot farmers, especially in developing countries. So-called eExtension services provide up-to-date, reliable information for small-plot farmers, but struggle to collect the plot-specific crop health information on which to base personalized recommendations. This thesis addresses this issue using novel ideas in remote sensing system understanding, image processing, and geospatial workflow to develop the Airborne Monitoring System for In-Season Agriculture (AMSISA) and make remotely sensed crop health data accessible and useful for small-plot farmers. A simulation of platform performance characteristics shows that manned aircraft are the better aerial remote sensing platform, given current performance and regulatory realities. The time-series aerial remote sensing (TSARS) approach allows a reduction in spatial resolution to dramatically reduce survey costs and enable frequent updates for better monitoring performance. This allowance for plot-resolution (but not finer) data and minimal cost per hectare over a large area leads to the development of a model to optimize the data collected per plot based on survey altitude, heading, and camera properties. Imagery collected using a custom camera setup aboard a manned aircraft in Maharashtra, India is used to test and verify the models. Surveying at a spatial resolution near the size of farm plots on the ground requires precise registration of remotely sensed images to ensure accurate crop reflectance measurements. Current and novel multi-modal image registration techniques are tested and found to be inadequate for this application. Instead, a technique using known fiducials is presented to achieve the required registration precision.