Skip to content

Set of Machine Learning Algorithms developed with the aim of determining health states of different types of crops

License

Notifications You must be signed in to change notification settings

RentadroneCL/AI-Agro

Repository files navigation

Discord

AI Agro

Remote sensing has as one of its objectives, to be able to provide useful information in the shortest possible time for decision-making. Therefore, it is considered a fundamental tool in precision agriculture, since it allows the monitoring of crops throughout the growing season, providing timely information as a diagnostic evaluation. This task must identify the factor that operates in a restrictive manner and decide, in a timely manner, on corrective agronomic intervention.

A promising approach to this is one that integrates data derived from temporal, mosaic, multispectral, and thermal imaging. Both processes allow us to obtain products such as: Thermal maps and Normalized vegetation index maps; These products allow us to identify stress zones which serve as support in agricultural management tasks.

That is why our objective is to develop an open-source platform, distributed on a GitHub platform, that is capable of generating local calculations and mapping (plant by plant) of most important vegetation indices, through the processing of images taken with UAV.

To learn all about it, head over to the extensive documentation.

Software Features (To Do List):

  • Open Source, distributed on GitHub platform
  • Able to map the state of health in different types of crops visible in multispectral photographs taken with drones, allowing the calculation of the main types of vegetation indices (NVDI, GNDVI, NDRE, LCI, OSAVI, etc.)
  • At a minimum, it must be able to process JPG and TIFF (Multispectral Radiometric) images.
  • Possibility of generating multispectral orthomosaics for each band and for each vegetation index. In addition, it must be possible to extract pixel intensity values in case calculations of geolocated variables are required.
  • Be able to perform batch processes with batches of photo files.
  • Online and Local Multiplatform Operation.
  • Generate KMZ maps, using the GPS information in metadata of the photos
  • Have a module for generating statistical reports regarding the number and types of problems found in photographs.

Multispectral band wavelengths available.

Today the sensors of the cameras on board UAV can capture spectral images in the wavelengths of red, red- edge, near infrared and thermal (Table Nº1).

Table Nº1: Multispectral bands

Band Wavelength
Blue 450 nm
Green 560 nm
Red 650 nm
Red Edge 730 nm
Near infrared 840 nm

Vegetation index calculations

The following spectral index can be generated from these lengths (Table Nº2).

Table Nº2: Spectral index generated from the available wavelengths of camera on board UAV.

Index Equation
Normalised Difference Index NDVI = ( Rnir - Rr)/(Rnir+Rr)
Green Normalized Difference Vegetation Index GNDVI = (Rnir - Rgreen)/(Rnir + Rgreen)
Normalised Difference Red Edge NDRE = (Rnir - Red edge)/ (Red edge + NIR)
Leaf Chlorophyll Index LCI = (Rnir - Red edge)/(Rnir + Red)
Optimized Soil Adjusted Vegetation Index OSAVI = (Nir - Red)/(Nir+Red+0.16)

Defining plant health status labels

NDVI 1 NDVI 1 < NDVI 2
Rank Description Description
-1 to 0 Water, Bare Soils Water, Bare Soils
0 to 0,15 Soils with sparse, sparse vegetation or crops in the initial stage of development (sprouting) Poor vigor, weak plants
0,15 to 0,30 Plants in intermediate stage of development (leaf production) Bad leaf / flower ratio
0,30 to 0,45 Plants in intermediate stage of development (leaf production) Bad flower / fruit ratio; fruits with low sugar content, lack of color in the fruits, fruits of low caliber
0,45 to 0,60 Plants in the adult stage or phase (fruit production) Bad flower / fruit ratio; fruits with low sugar content, lack of color in the fruits, fruits of low caliber
0,60 to >0,80 Plants in the adult stage or stage (Fruit maturity) Bad flower / fruit ratio; fruits with low sugar content, lack of color in the fruits, fruits of low caliber

Limitations of this solution

  1. The multispectral orthomosaics have to be built before using the tool
  2. Process the RGB and multispectral bands separately
  3. Must know the format of the bands you will be using and the metadata of each image (tiff, GeoTiff)
  4. At this moment, the methodology and support is only for Phantom 4 RTK Multispectral user.

Methodology

To complete the main objective we consider following diagram methodology (Image Nº1). Was proposed, which reflects the process of generating the information necessary for decision- making during the management of a production cycles of a crop in general.

Process_Diag

Several diagrams of sub- processes were also proposed. 1- To assess the growth status of plants. Image Nº2, NDVI multi-time series.

Diagrama2

Minimum Viable Product (MVD) for NDVI

MPV-NDVI

Contributing

Contributions are welcome and will be fully credited. We accept contributions via Pull Requests on GitHub.

Pull Request Checklist

Before sending your pull requests, make sure you followed this list.

Licenses

Everything in this repository is under a GNU General Public License v3.0.

All The data sources & datasets in this repository is under a CC-BY-4.0.