30 IRRIGATION AUSTRALIA CRC for IRRIGATION FUTURES The existence of spatial variability in the field means that the water requirements of different areas of a crop may not be the same. There is potential for saving water by variably applying water in response to real-time, site-specific irrigation requirement. however, real-time sensors need to be developed to measure real-time plant water stress at high spatial resolution. For cotton, the best indicator of water stress is plant growth. A cotton plant adds a new node to the main stem every three days, so the distance between these nodes on the main stem indicates the vegetative growth of the plant. Internode length is an attribute measured by agronomists to identify how moisture stressed a cotton crop is. However, measuring internode length is a tedious manual task, and it would WHEN AND WHErE TO WATEr? ASK THE PLANT, AuTOMATICALLy CRC for IRRIGATION FUTURES Cheryl McCarthy, CRC for Irrigation Futures be much better if the process could be automated. Potential of machine vision Machine vision, which can measure automatic internode length, is one possibility for doing this. Machine vision involves the extracting of useful information in a scene from a two- dimensional projection of the scene, such as an image captured by camera. A vision system has been designed to automatically measure internode length. The system features a camera enclosure that traverses the crop canopy. The enclosure comprises a video camera sitting behind a transparent panel, and uses the flexible upper main stem of the plant to firstly touch the plant against the transparent panel, and then smoothly and non-destructively move over the plant. When the main stem touches the transparent panel, it becomes a fixed object plane from which reliable two-dimensional geometry can be measured. Software algorithms have been written to take the image sequences collected by the camera and then systematically identify the main stem, plant branches, and then nodes, which are the intersection of the branches with the main stem. From a sequence of detected node positions, the distance between nodes, or internode distance, can be measured. Using the developed process, the system has automatically measured internode length with a standard error of 1.0 mm and correlation coefficient of 0.92. Calculations indicate that the algorithms can be performed in real time. A patent application has been lodged for this method and apparatus by the CRC’s company IF Technolofgies Pty Ltd. This research represents a step forward in machine vision of plants in the field environment. Typically, automatic machine vision measurement of plant parameters in the field is restricted to whole- plant characteristics, such as plant height or plant spacing. Publicly reported research for the automatic identification of subplant features such as stems and nodes is presently limited to seedlings and controlled glasshouse environments. The current research has demonstrated that a within-canopy camera enclosure is a suitable platform for automatic measurement of plant parameters in the field. Automatic internode length measurement has been successfully achieved. The sensor may potentially be used in a real-time application, such as a variable- rate irrigation machine, and could be implemented at relatively low cost. Figure 1. (top) Moving image capture apparatus; and (below) sample image from apparatus. Automatic conveyance of camera enclosure in a cotton crop. Cheryl McCarthy is doing her PhD with the CRC for Irrigation Futures. As part of this research she is looking at the potential of using machine vision to identify when and where to water plants. Cheryl’s is doing this research on cotton crops being grown on the Darling Downs in Queensland.