Intra-row weed detection in wheat at early growing stage using imaging system
Résumé
In the present agroecological context, the reduction of herbicide uses becomes a substantial issue. For weed control and site specific weed management the development of new technologies is required to identify individual plant. Although many crop/weed discrimination methods have been developed, the automatic intra-row weed detection is still a challenge, particularly in wheat. An imaging system is developed to detect black grass in wheat at early growing stage. The technique mimics the method used by a human expert in such conditions. Thus, the hypocotyl diameter was used as distinguishing criterion. For experiments, the seeds of black grass and wheat were sown in rectangular pots. Two planting protocols were carried out to reproduce the emergence difference of species as observed in actual field conditions: 1) black grass and wheat were planted in the crop line at the same time; 2) black grass were planted in the crop line seven days after wheat. The imaging system consisted in a multispectral camera placed at 1.20 m above the pot soil surface. The view axis was tilted at 45° so that the side view of the crop row was captured in the image. Approximately 100 images were acquired at various times from plant emergence (BBCH-stage 11) to three leaf stage (BBCH-stage 13) of the wheat plant. The first step of image processing consisted in computing a color vegetation index (Excess Green Index) and thresholding the pixel values to provide a binary image of the plants without soil background. In the second step, various segmentation and filtering techniques were assessed to detect automatically the base of plant stems. Depending on the stem width, each plant was finally classified into wheat or black grass. Results are analyzed and discussed looking at the potential of this method regarding not only the plant species but also experimental conditions.
