DINA Workshop on Statistics and Image Analysis

Thursday November 16, 2000, at Aalborg University.

Estimation of plant density and weed infestation

from aerial red/near-infrared images.

Bent S. Bennedsen

AgroTechnology, The Royal Veterinary and Agricultural University

The research reported in this paper investigated the feasibility of studying a canopy from aerial red-near infrared images. The reflectance values, corresponding to different grey levels on the images, depend on species, growth stage, conditions of the culture and indirectly on soil composition through both last parameters. The objective of this study was to determine the accuracy to which it is possible to identify parameters from aerial photographs in order to predict the state of the culture for later yield predictions.

The equipment used for data acquisition was an R-NIR aerial photography system, which had been specified and purchased by the Danish Research Centre Bygholm in 1992. It consists of three CCD cameras which are sensible in the near-infrared field up to 1050 nm. Each camera was provided with a low-pass filter with cut-offs at 850 (Camera 1), 760 (Camera 2) and 660 (Camera 3) nm respectively. Three images are recorded, one from each camera, approximately synchronously. The system is fixed on the door of an aircraft (a Cesna 172).

Two fields at the Royal Veterinary and Agricultural University in Taastrup were selected for the project: An experimental field of spring wheat of small size (44mx40m) with different levels of nitrogenous fertilisers, seeding density, seeding patterns and varieties. The field was divided in 96 subunits of 88x1,4m and each was a particular combination of these four variables. A "normal" field of winter wheat (60x230m) on which variations of soil composition have been detected and measured from sampling and laboratory analysis

Three missions were flown during spring and summer 2000 Images were taken by flying vertically over the crop canopy at a height varying from 240 to 420 m depending on the size of the field. Image processing method consisted in calculating the NDVI (Normalised Difference Vegetation Index) which represents the aerial chlorophyllian vegetation. For pictures taken at first stages of growth the SAVI (Soil Adjusted Vegetation Index) was chosen instead of NDVI..

Results shows an up to 75% correlation with plant density during the early stages of growth (18^th May), whereas this correlation decreases significantly during the vegetation period. The correlation can be utilised to estimate weed infestation during springtime. Observations, mainly from the second field confirms previous findings, that good correlation can be obtained with crop nitrogen level. However, one of the main results is, that although varying image intensity can be successfully related to a number of crop parameters, best results are obtained when all other parameters are equal. This calls for further research to increase the level of inputs.