The mobile platform is currently being used to capture the following measurements at the DPIRD frost research site at Dale;
- Greenseeker® for taking Normalized Difference Vegetation Index (NDVI) readings
- Digital camera for assessment of percentage groundcover
- Plant height/biomass sensor (currently in prototype)
To take all three readings from 400m x 5m plots takes one hour and directly uploads the data to datasheets, ready for analysis.
An earlier prototype with an electric assist motor, which was ridden by the operator was used successfully for several years the Managed Environment Facility in Merredin using both the Greenseeker® and digital camera. However keeping the bike at a constant speed was an issue with varying soil compaction. The current version is pushed with speed control no longer a problem, although depending on the terrain the power assist could, be used especially on long plots.
The NDVI sensor was designed to be held 81-122cm over the target canopy and the width of the sensor is a constant 61cm, independent of height, according to the manufacturer. The Greenseeker® head has been positioned in the centre of the platform parallel to the rows and depending on row width may be sensing the central two or three rows. As per normal sampling techniques, readings are not taken within the first or last metre of plot.
The advantage of using a platform is that the readings are more consistent as the height, position within the plot and orientation of the sensor head is fixed.
A consideration when using the Greenseeker® for a prolonged period by hand, is fatigue and back strain, even when using the harness provided. The height of the reading and the orientation of the head may change as fatigue sets in and both of these can affect readings. Subsequent assessments may also not be from the same area as previous assessments.
For more information refer to D Martin et al 2012 'Laboratory evaluation of the GreenSeeker® hand-held optical sensor to variations in orientation and height above canopy', in the links section.
When using the camera mounted on the platform, the photographs are from a consistent height and the operator, does not have to stop to take a photo or position themselves within the crop.
The disadvantage of opting for manual operation of the camera include inconsistent height of photograph, fatigue and muscle soreness from extending the arm to take a photograph which leads to more error and poor quality images.
|Plot||Time||NDVI||Encoder reading||Groundcover (%)|
Digital images for estimation of groundcover
There are several software programs available for processing photographs to estimate the percentage of groundcover.
The programs trialed in the frost project include;
- Adobe® Photoshop®
- Canopy Cover
This program needed a few attempts to become familiar with the procedure but when mastered it was quite easy to use. Each occasion a set of photographs are taken, settings need to be manipulated. Mike found that to batch process a few hundred photographs takes 30 minutes.
Developed by CSIRO requires minimal input. However to batch process a few hundred photographs may take 3-4 hours. The program also needed a dedicated computer, as once processing was completed the results would be pasted to the clipboard.
Canopeo is a rapid and accurate green canopy cover measurement tool. You can use this app to quantify the percent canopy cover of live green vegetation for any agricultural crop, turf or grassland based on downward-facing photos taken with your mobile device. The processing time for a batch of photographs was similar to Photoshop®.
The height/biomass sensor is basically a swing made from PVC pipe, positioned centrally on the platform to sample the middle two rows and when moved over the plot, the flap rises or falls according to biomass.
The top of the flap is, attached to the shaft of a rotary encoder, an optoelectronic device and as the shaft rotates, it breaks a light beam enabling the encoder to sense its position relative to the vertical axis (see below for more detailed information on this). The principal is similar to the rising plate meter, which has been in use for over forty years to measure biomass of pastures. Refer to C Stockdale 1984 'Evaluation of techniques for estimating the yield of irrigated pastures intensively grazed by dairy cows' in the links section of this page.
Teething issues: After a few problems with programming and hardware errors, Mike was able to calibrate plant heights by taking a few measurements within each plot, using a square polystyrene plate placed over two rows of the crop in two places per plot and recording heights. For the results, refer to Plant biomass and height calibration or in the links section of this page.
With the use of a 600 pulse per revolution rotary encoder, an Arduino board or an SD card with appropriate coding, the position of the swing is recorded. When the Arduino is first switched on it is imperative that the swing is vertical (right angle to the crop). The encoder will record this as a home position and read 0. At its maximum height (parallel to the ground) the encoder will read 600.
There are a few limitations including:
- Many crops are sown with furrows to harvest water. This leaves clods at the edge of the furrow and without harrowing or rolling, if the swing was to be at ground level readings would be influenced by the clods. It is difficult to record accurate heights before the crop reaches about 100 mm. For this reason the base of the swing is about 75mm above the ground.
- The PVC frame may be affected by strong winds.
- Best use of the height biomass sensor is prior to plant maturity. This is because the plants become more rigid. However it may be possible to use different weighted gates to continue sampling.
- Bumpy terrain can often cause false triggering. By coding with a delay and a switch de-bounce routine has over come this. False triggering is now more likely caused by inadvertently touching the trigger while turning at the end of an experimental plot.
Some practical tips include:
- The frost experimental plots are in rows/ranges. It is more efficient to work in a serpentine pattern. To minimise upload error, before the first plot of run one, Mike stops and takes a reading on bare ground which will give a low encoder reading. This is then repeated before the start of the next run. This processes helps to limit incorrect data.
- The number of readings per plot is predefined by code. When taking readings the operator should push at a constant speed. A few practice plots could be read first to ensure that readings finish within the plot. Walking too slowly will result in only a short section of plot being read. Readings could be aligned to a length of plot but again walking pace influences the number of readings. Fast walk, fewer readings, a slower walk gives many readings.
When taking readings, plots of one species of similar development stage and seeding rate should only be compared unless calibration cuts are taken for each species. For a simple explanation of rotary encoders and the example code Mike modified refer to PJRC's Encoder library page. Contact Mike Baker directly to discuss customisation of the code for the purpose of the mobile platform.
- The ‘Mikemobile’ was constructed from an unwanted bicycle and steel offcuts within approximately week in the experimental workshop.
- The electronic component has been a hobby interest of Mike's, taking a few days to construct.
- The cost of metal for construction would be approximately a couple of hundred dollars.
- The cost of electronic components would be several hundred dollars.
- An electric assist bike motor, battery and speed control would cost another $600-1000.
- A good design engineer could make the assembly more portable and easier to handle, as at present it is heavy and cumbersome for one person to load/unload from a vehicle and assemble/dissemble (currently a two person job). A lighter bike frame would be a great start to addressing these issues.
Learning programming to control the Arduino was time consuming. The parts required for this include:
- Arduino Uno Board
- SD card reader
- Switches various
- Arduino jumper wires
- Incremental Rotary Encoder 600ppr 400ppr is okay but resolution is lower
- 5 volt Arduino active buzzer
- LEDS, resistors
- 12 volt battery to power Arduino
- Waterproof electrical box to house all components.
- Computer with Arduino installed with Libraries included in code (contact Mike Baker to discuss his customisation of the code). The version currently being used within the frost project is Arduino 1.8.3