Development of an integrated approach for weed detection in cotton, for site specific weed management

The ability to properly control and eliminate weed infestation in cotton fields can be a crucial factor for the crop's profitability. Site specific weed management has long been marked as the next step in precision agriculture and the tools of remote sensing and image processing stands in the front line of this discipline. Several researches that have conducted trials in this field showed encouraging results for distinguishing weeds from the crop, but no mechanism has been implemented yet in the agriculture routine. This fact has driven us to develop a low cost, yet effective, method for weed map preparation which later on will be used for on-the-go herbicide application. A new approach is suggested which includes combining information from a hydro-thermal model into image processing algorithms. This information consists of the expected number of plants per m2, leaves per plant and other inouts. The suggested system consists of a simple RGB camera, mounted on a tractor's sprayer boom and connected to an on-board computer. A video file is recorded with the camera at a rate of 7.5 frames per second. The pictures are then being extracted from this file and analyzed in a MatLab™ environment. The process involves applying different vegetation indices using the RGB channels. Object separation is based on morphological features and compared to incoming information from the hydro-thermal model and data from previous years. The result of this process is a 'weed map' calibrated with DGPS data that had been collected during the image acquisition stage. The information achieved each year will also be used for continuous update and amendment of the hydro-thermal model.