Using remotely controlled platform to acquire low-altitude imagery for grain crop mapping

Using remotely controlled platform to acquire low-altitude imagery for grain crop mapping

[Abstract]Agricultural crops exhibit within-field spatial variation. This variation partly results from relevant bio-physical and environmental factors that influence the
crop during the growing season. The plant integrates the effects of nutrition, water, pests and disease, and displays the results in the foliage. Remote sensing techniques allow the foliage to be monitored and the crop status to be assessed.

While the use of conventional remote sensing systems has found many applications in agriculture, it is constrained by a number of issues and problems related to spatial resolution, repeat cycle, minimum area acquired, timeliness of data, etc. Thus, this research explores the potential of developing and assessing low-cost sensing technologies to overcome these limitations. The specific
objectives were to: a) identify, evaluate, and analyse the different options for a low-cost low-altitude (LCLA) remote sensing system that has potential for precision agriculture, b) develop a LCLA remote sensing system that is appropriate for use in mapping selected crop attributes (i.e. grain protein, yield, maturity and crop type), and c) evaluate the accuracy of classification and prediction of the cereal crop attributes.

A low-cost sensor system was developed that incorporated two consumer digital still cameras. One camera captured the colour portion of the spectrum, while the other one (with the addition of a band-pass filter) captured the near
infrared light. Both cameras were modified to be remotely triggered and externally powered. This sensor arrangement utilised 1.0 megapixel cameras in the earlier investigations and then 5.0 megapixel cameras in most recent missions. The sensors were equally well suited to mounting on a remotely controlled aircraft or suspended beneath a helium balloon.

Various approaches were taken to determine and evaluate the relationships between imagery and crop attributes. Statistical methods included the use of correlation and discriminant function analysis, along with partial least squares regression. Image analysis techniques included the use of both pixel-based (supervised approach) and object-orientated (multi-resolution segmentation)
classifications.

The results showed that low-cost low-altitude remote sensing systems (incorporating consumer digital cameras with helium balloons or remotely controlled aircraft) have great capacity to quantify variability in cereal grain
crops. Excellent relationships were found between the ‘at-harvest’ yield (R2=0.902) and protein content (R2=0.660) of wheat using a single image recorded at flowering. Partial least squares regression, using the crossvalidated
approach, produced a stronger relationship with a prediction accuracy of 94.2% for yield and 88.5% for protein. This relationship exceeded all other studies reported in the literature.

The same LCLA system has also accurately discriminated (using statistical methods) between: a) different nutrition levels in a wheat crop with 75.6% of the cases correctly classified, and b) between different cereal grain species (with differing nutrition levels) with 86.3% accuracy. These classification accuracies are comparable with, or exceeding other more expensive and/or complicated methods. Attempting to discriminate using image analysis
procedures, the pixel-based methods yielded an overall accuracy of 65.9% when classifying cereal grain crop species comprising of nine classes. When merged to six classes, the accuracy improved to 82.1%. Using an objectorientated approach has improved the overall accuracy to 81.0% for the ninecategory classification. This study also demonstrated LCLA’s ability to assess
the various growth stages of a barley crop prior to maturity with 83.5% of cases correctly classified.

This study concluded that it is feasible to accurately assess selected cereal grain crop attributes using low-cost consumer technologies. The accuracies achieved
using this system were comparable with, or exceeded, other reported studies that used more complicated and expensive sampling systems. Further work is needed to continue refining the initial work on a fully autonomous unmanned
aerial vehicle (UAV) started in the later part of this study, to extend the use of the LCLA system into broader scale applications.