Development and Evaluation of a Real Time Site-Specific Inter-Row Weed Management System

ABSTRACT- A real-time, site-specific, machine-vision based, inter-row patch herbicide application system was developed and evaluated. The image resolution was 640 × 480 pixels covering a total area of 350 mm x 240 mm of a field composed of four quadrants of 350 mm x 60 mm each. The image frames were processed by LabView® and MatLab®. The developed algorithm, based on weed coverage ratio and segmentation method for separating soil from plants, was chosen to be 2G-R-B. The unit was lab-tested to evaluate the accuracy of the sprayer at three travel speeds of 2, 4 and 6 km h-1. The effect of travel speed on spraying delay was not significant when using LabView®, while it was significant when MatLab® was used for image processing. A binary analysis algorithm showed that misclassification rates for plant (MCRP) and soil (MCRS) were (2.4±1.3)% [mean±SD] and (2.8±0.7)%, respectively and correct classification rates for plants (CCRp) and soil (CCRs) were (96 ±1.1)% and (93.8±1.3)%, respectively. During the field test, three herbicide application treatments consisting of target patch spraying, conventional (uniform) spraying and no spraying (control) were compared at three levels of weed coverage (low, medium and high). In the patch spraying treatment, a spray mixture of 2-4-D (0.35%) was sprayed only over the weed patches, while the conventional treatment plots received uniform and continuous applications of the same mixture. The result showed that the herbicide used in each patch spraying plot was proportional to the weed coverage level. On average, the target application resulted in 75% less herbicide consumption compared to the conventional application, while the patch spraying application was as effective as the conventional application for the eradication of weeds.