An automated sensing system, hardware and software, was developed for spot-application of herbicide with 6.1 m boom automated prototype sprayer. The sprayer was consisted of 8-channel computerized variable rate controller (VRC), Dickey John Land Manager II controller, pocket personal computer, solenoid valves, servo valve, flow meter and nozzles and mounted on all-terrain vehicle (ATV). The sprayer boom was divided into eight equal sections. Four μEye-1220SE cameras were installed on a separate boom in front of ATV and connected with a ruggedized PC through USB cables. Custom software was developed in C⁺⁺ programming language and installed in PC to acquire images from cameras and process in real-time to calculate green pixels for weed detection in each image. Clear green contrast between weed and plants was used to develop a robust and effective algorithm including converting the 24-bit Red-Green-Blue (RGB) images to normalized green ratio images (255*G/(R+G+B)), segmentation by threshold values, and counting over-threshold pixels. Custom software was capable of processing images to differentiate weeds from plants in real-time and send ON/OFF signals to VRC to open the valve in a specific section of the boom where the weeds have been detected.

Green ratio (GR) algorithm was compared with excessive green index (EXG) algorithm. Both algorithms were evaluated with visual inspected images taken from different fields. The GR algorithm showed reliable accuracy (> 99 %) under different outdoor light conditions (> 500 lux). The GR algorithm showed better accuracy than EXG algorithm due to the noise under shade and low light condition. The sensing system performed well for accurate weed detection and sending signal to VRC for spraying at right targets up to 8 km/h ground speed. Kaolin dye and water sensitive paper were used to test the system in different fields. Detailed results are presented in the paper.