The laboratory's work is organised around the use of imaging in agriculture, which can be used either for checkups on the state of agricultural machinery (Theme 1) or for the characterisation of plants (Theme 2).
Mechanically simple, centrifugal spreading counts for around 85% of the market in fertiliser spreading techniques due to its sizeable economic advantages. This technique still reveals complex physical phenomena, due in part to the high number of parameters necessary for a classic centrifugal spreader.
Expertise in spreading operations necessitates:
S. Villette (2006) : Estimation de vitesse par analyse d'images acquises en filé : application à la caractérisation de la distribution centrifuge de granulés d'engrais. (Estimation of speed via analysis of high-speed imagery: as applied to the characterisation of centrifugal spreading of fertiliser granules)
A multispectral imager was created in the laboratory to map cultivated plots of land, to obtain maps of weed infestation.
This imager (a camera with 4 filters: B, G, R, IR), was attached to a small model aeroplane (Thesis J.B. Vioix – 2004). This prototype is undergoing optimisation so as to geolocate the images thus taken.
The prototype was modified to be a portable terrain imager. The tests were done on plots of land belonging to the INRA at Epoisses, in conjunction with the UMR BGA (INRA/AgroSup Dijon).
Thesis JB. Vioix (2004): Conception et réalisation d'un dispositif d'imagerie multispectrale embarqué : du capteur aux traitements pour la détection d'adventices. (Design and creation of an on-board multispectral imager: from the sensor to methods of weed detection)
Also dedicated to creating maps of weed infestation in a given plot of land, this work on the comprehension of the spectral and microscopic properties of weeds attempts to classify or identify the main groups of plants via their spectral signatures.
This work on plant spectroscopy is done in conjunction with Le2i (F. Marzani) and the UMR BGA (L. Assemat – INRA and J.-P. Guillemin – Enesad). It adds to the work on the identification of plants based on their spatial properties within a given cultivated plot of land.
Concerned with agricultural processes respectful of the environment, we became interested in the development of an imaging system for precision spreading. The idea was to reduce doses of herbicides sprayed on cultivated parcels of land (cereal crops).
A monochromatic camera is loaded onto the front of a tractor. The image is analysed in real-time by a computer to identify the vegetation from the rest of the surroundings and then to separate out weeds from crops. This process is carried out by using a Gabor filter which determines the volume of the crop. Once the infestation map has been drawn up each solenoid valve can be activated at the right moment, during the right period.
Thesis by J. Bossu (French document in PDF format - 7,4 Mb): "Application à la réalisation d'un système de vision pour une pulvérisation spécifique en temps réel" (Application of an imaging system for precision spraying in real time) - University of Burgundy - defended on 4 December 2007.
The reliability of the algorithms (Gabor Filter, Hough transform, Wavelet transform) developed in the laboratory to identify the weed crops in the images taken was never proved.
One solution is to develop a spatial and spectral model of an agronomic area with which to test these algorithms.
The advantage of this method is to be able to perfectly control the initial parameters of the area (type of crop, frequency of inter and intra rows, infestation count) but also to envisage all the images shot on this model parcel of land by using a virtual camera whose parameters can also be controlled (rotation, focal, size of CDD...).