Row and water front detection from UAV thermal-infrared imagery for furrow irrigation monitoring

Water efficiency in furrow irrigation has been improved by the introduction of feed-back sensing systems, which help inform the decision on when to cut the water off for optimal use, but typically only a limited number of furrows can be monitored using existing sensors. The aim of this research is to develop automatic machine vision algorithms for UAV (also known as Remotely Piloted Aircraft, or RPA) thermal imagery, collected as the UAV traverses overhead of a cotton crop, to monitor furrow irrigation progress of large areas of a field. An algorithm was developed for overhead thermal imagery of a cotton field with high canopy closure. A test flight with a < 2kg multirotor UAV was performed in late February, 2016 to assess the accuracy of the algorithm. It was found that at lower sensing heights (20 m), most water fronts were being detected, with a significant drop in performance at the higher altitude of 30 m. The algorithm also estimated the row direction and spacing relative to the camera, and used the estimates to calculate the row number for each detected front. The average error in water front position estimation was between 1.3 and 2 m which is well within limits for practical irrigation management. The water stream was found to be visually discernible in all crop rows captured in the overhead thermal imagery, despite the water stream not being visually discernible in overhead color imagery captured in the same UAV flights, due to the level of canopy closure