Results obtained from 155 multispectral images taken by drone over a 18ha area.
Below are the different stages of a service aimed at mapping each plant according to its plant vigor (according to the NDVI index). Generation of a multispectral orthophoto, automatic counting by deep learning, segmentation and assignment of NDVI value for each plantation.
As illustrated below with Agisoft Metashape, first step consist in 3D modeling the entire site and then export a multilayer orthophoto with 12 spectral channels (in accordance with the 12 bands generated natively by the A7R4 Quad for each shooting).
Through this use case, only just 4 channels are useful: red, green, blue and near infrared. This for the subsequent built of a conventional RGB orthophoto (for visual comfort) and the NDVI index calculation as illustrated below from the red and near infrared channels.
Thanks to the high resolution of this Agrowing sensor, the previously built orthophoto (GSD2cm with a 115m height flight) offer a very high level of details which perfectly reproduce the singular shape of almond trees. In this context, detection by deep learning based on color & shape can be considered.
Once the almond trees are perfectly segmented with respect to each other, a NDVI value can then be assigned locally for each tree.
In addition to its 12 spectral channels dedicated to remote sensing indexes customization, the high resolution of the A7R4 Quad sensor make it possible to generate extremely well detailed NDVI maps but also to enrich them with automatic detection/counting: an unique advantage that meets perfectly to ecological issues and precision farming needs.