No abstract

This paper outlines how light Unmanned Aerial Vehicles (UAV) can be used in remote sensing for precision farming. It focuses on the combination of simple digital photographic cameras with spectral filters, designed to provide multispectral images in the visible and near-infrared domains. In 2005, these instruments were fitted to powered glider and parachute, and flown at six dates staggered over the crop season. We monitored ten varieties of wheat, grown in trial micro-plots in the South-West of France. For each date, we acquired multiple views in four spectral bands corresponding to blue, green, red, and near-infrared. We then performed accurate corrections of image vignetting, geometric distortions, and radiometric bidirectional effects. Afterwards, we derived for each experimental micro-plot several vegetation indexes relevant for vegetation analyses. Finally, we sought relationships between these indexes and field-measured biophysical parameters, both generic and date-specific. Therefore, we established a robust and stable generic relationship between, in one hand, leaf area index and NDVI and, in the other hand, nitrogen uptake and GNDVI. Due to a high amount of noise in the data, it was not possible to obtain a more accurate model for each date independently. A validation protocol showed that we could expect a precision level of 15% in the biophysical parameters estimation while using these relationships.

A semi-automatic system was developed to monitor micro-plots of wheat cultivars in field conditions for phenotyping. The system is based on a hyperspectral radiometer and 2 RGB cameras observing the canopy from ~1.5 m distance to the top of the canopy. The system allows measurement from both nadir and oblique views inclined at 57.5° zenith angle perpendicularly to the row direction. The system is fixed to a horizontal beam supported by a tractor that moves along the micro-plots. About 100 micro-plots per hour were sampled by the system, the data being automatically collected and registered thanks to a centimetre precision geo-location. The green fraction (GF, the fraction of green area per unit ground area as seen from a given direction) was derived from the images with an automatic segmentation process and the reflectance spectra recorded by the radiometers were transformed into vegetation indices (VI) such as MCARI2 and MTCI. Results showed that MCARI2 is a good proxy of the GF, the MTCI as observed from 57° inclination is expected to be mainly sensitive to leaf chlorophyll pigments. The frequent measurements achieved allowed a good description of the dynamics of each micro-plot along the growth cycle. It is characterised by two periods: the first period corresponding to the vegetative stages exhibits a small rate of change of VI with time; followed by the senescence period characterised by a high rate of change. The dynamics were simply described by a bilinear model with its parameters providing high throughput metrics (HTM). A variance analysis achieved over these HTMs showed that several HTMs were highly heritable, particularly those corresponding to MCARI2 as observed from nadir, and those corresponding to the first period. Potentials of such semi-automatic measurement system are discussed for in field phenotyping applications.

Crop density is a key agronomical trait used to manage wheat crops and estimate yield. Visual counting of plants in the field is currently the most common method used. However, it is tedious and time consuming. The main objective of this work is to develop a machine vision based method to automate the density survey of wheat at early stages. RGB images taken with a high resolution RGB camera are classified to identify the green pixels corresponding to the plants. Crop rows are extracted and the connected components (objects) are identified. A neural network is then trained to estimate the number of plants in the objects using the object features. The method was evaluated over three experiments showing contrasted conditions with sowing densities ranging from 100 to 600 seeds⋅m-2. Results demonstrate that the density is accurately estimated with an average relative error of 12%. The pipeline developed here provides an efficient and accurate estimate of wheat plant density at early stages.

No abstract