Group Actions
Engagement
  • Author Author: gpolder
  • Date Created: Date Created
  • Views 4078 views
  • Likes 17 likes
  • Comments 15 comments
Related
Recommended

[Pi IoT] Plant Health Camera #1 - Application

gpolder

Introduction

Multispectral images including red and near-infrared bands have proved their efficiency for vegetation-soil discrimination and agricultural monitoring in remote sensing applications. But they remain rarely used in ground and UAV imagery, due to a limited availibility of adequate 2D imaging devices.

I was very happy to see that the new 8 Mpixel Pi Camera and the Pi Noir Camera were added to the kit of this challenge.

In the past I did an extensive test on the spectral properties of the old Pi Noir camera (Pi NoIR and Catch Santa Challenge - Review ). Now with the new sensor, with higher resolution and sensitivity I would rather like to use them to make a camera system for plant health analysis. Two approaches can be used: first obtain simultaneously the near-infrared and blue bands from the Pi Noir Camera. This can be done using the infra blue filter which was provided with the first Pi Noir camera (is this still the case?). NDVI values can be obtained obtained from the Pi Noir camera and can be compared with reference values for a set of soil and vegetation luminance spectra. Second approach is to use both cameras and use the NIR band from the Pi Noir and the red band from the Pi Camera. The images can be overplayed using image processing software like OpenCV, and the NDVI can be calculated.

The expected quality of the images are sufficient to obtain NDVI bands which can now be acquired with high spatial resolution, opening new opportunities for crop monitoring applications.

 

Plant Health measurements

The Normalized Difference Vegetation Index (NDVI) is a numerical indicator that uses the visible and near-infrared bands of the electromagnetic spectrum, and is adopted to analyze remote sensing measurements and assess whether the target being observed contains live green vegetation or not. NDVI has found a wide application in vegetative studies as it has been used to estimate crop yields, pasture performance, and rangeland carrying capacities among others. It is often directly related to other ground parameters such as percent of ground cover, photosynthetic activity of the plant, surface water, leaf area index and the amount of biomass. NDVI was first used in 1973 by Rouse et al. from the Remote Sensing Centre of Texas A&M University. Generally, healthy vegetation will absorb most of the visible light that falls on it, and reflects a large portion of the near-infrared light. Unhealthy or sparse vegetation reflects more visible light and less near-infrared light. Bare soils on the other hand reflect moderately in both the red and infrared portion of the electromagnetic spectrum (Holme et al 1987).

Since we know the behavior of plants across the electromagnetic spectrum, we can derive NDVI information by focusing on the satellite bands that are most sensitive to vegetation information (near-infrared and red). The bigger the difference therefore between the near-infrared and the red reflectance, the more vegetation there has to be.

 

The NDVI algorithm subtracts the red reflectance values from the near-infrared and divides it by the sum of near-infrared and red bands.

 

NDVI= (NIR-RED) / (NIR+RED)

 

This formulation allows us to cope with the fact that two identical patches of vegetation could have different values if one were, for example in bright sunshine, and another under a cloudy sky. The bright pixels would all have larger values, and therefore a

larger absolute difference between the bands. This is avoided by dividing by the sum of the reflectances.

Theoretically, NDVI values are represented as a ratio ranging in value from -1 to 1 but in practice extreme negative values represent water, values around zero represent bare soil and values over 6 represent dense green vegetation.

(Also see: https://en.wikipedia.org/wiki/Normalized_Difference_Vegetation_Index )

 

Here is an example from (https://publiclab.org/wiki/near-infrared-camera )

 

image

 

 

Building an Raspberry Pi based NDVI Camera

Two approaches can be used:

  1. It is possible to capture all the information needed to compute NDVI using only the Pi Noir camera. If a filter is added that passes NIR and blocks only red light, then the red channel will record mostly NIR light. The blue channel which will record mostly blue light (some NIR light will also be captured in each channel) can be used to represent wavelengths that are absorbed by plants.  This can be done using the infra blue filter which was provided with the first Pi Noir camera (is this still the case?). NDVI values can be obtained obtained from the Pi Noir camera and can be compared with reference values for a set of soil and vegetation luminance spectra.
  2. Second approach is to use both cameras and use the NIR band from the Pi Noir and the red band from the Pi Camera. The images can be overlaid using image processing software like OpenCV (OpenCV | OpenCV ), using image registration functions (OpenCV: Image Registration ) and the NDVI can be calculated. Here is an example from (Automatic Optical and Infrared Image Registration for Plant Water Stress Sensing | InTechOpen)

image

Results of SIFT based registration algorithm (no matching key-points) (a) 5906 keypoints found in optical image, (b) 447 keypoints found in IR image, (c) 5666 keypoints found in optical image, (d) 605 keypoints found in IR image

 

The expected quality of the images are sufficient to obtain NDVI bands which can now be acquired with high spatial resolution, opening new opportunities for crop monitoring applications.

 

Physical build

Regarding the provided kit, I'm planning to use the Pi 3 as main computer with one camera attached. I expect that just one camera can be connected to a Pi, so the Pi B+ wil be used for the second camera. The Pi's will communicate over ethernet.

The touch screen will be used as user interface and also show the resulting NDVI images and numeric results.

The sense hat will be an valuable addition since during plant measurements the humidity and temperature are important parameters for instance in Greenhouses. Same for the enOcean sensors.

 

About me: I'm currently researcher machine vision and plant phenotyping at the Wageningen University in the Netherlands. Started as an electronic engineer. In 2004 I got a PhD. from Delft University of Technology on Spectral imaging for measuring biochemicals in plant material. From 2004 working on machine-vision and robotics projects focused on agricultural research.  I have been an electronics hobbyist and radio amateur for more than 30 years and I work on Wireless, DSP, SDR and embedded products.

 

 

References

https://en.wikipedia.org/wiki/Normalized_Difference_Vegetation_Index

https://publiclab.org/wiki/ndvi-plots-ir-kit

Pi NoIR and Catch Santa Challenge - Review

https://hal.archives-ouvertes.fr/hal-00648439/document

http://w3.avignon.inra.fr/valeri/documents/JonckheereAFM2003Accepted.pdf

OpenCV | OpenCV

Automatic Optical and Infrared Image Registration for Plant Water Stress Sensing | InTechOpen