DEVELOPMENT OF A MACHINE VISION BASED WEED (GOLDENROD) DETECTION SYSTEM FOR SPOT-APPLICATION OF HERBICIDES IN WILD BLUEBERRY CROPPING SYSTEM
Abstract
Wild blueberry crop yields are dependent on heavy agrochemical applications to control weeds competing with crop. Goldenrod is a creeping herbaceous perennial weed that occurred in more than 90% of wild blueberry fields surveyed in Nova Scotia. The objective of this study was to develop and evaluate a graphical user interface based goldenrod detection system using colour concurrence matrices as image processing algorithm and machine learning procedures for spot-application of Callisto® herbicide. The performance of developed goldenrod detection system was tested and evaluated in four wild blueberry fields. Results of laboratory evaluation suggested that developed colour co-occurrence matrices algorithm with a back-propagation classifier has ability to target the goldenrod with an accuracy of 97%. Optimum parameter selection suggested that intensity levels of 256 and a unit image size of 128 × 128 pixels can help to minimize the processing time without compromising the classification accuracy for real-time applications. Results of classifiers development showed that back-propagation artificial neural network classifier performed better than statistical quadratic classifier to classify goldenrod for training and test datasets. Field evaluation results confirmed the higher accuracy of artificial neural network classifier compared to statistical quadratic counterpart. The savings with colour occurrence matrices and artificial neural network classifier were in the range of 32% to 65% depending upon the goldenrod coverage within selected field tracks. This study can help to apply Callisto® site specifically for goldenrod control, thereby allowing the producers to apply agrochemicals in an economic and environment friendly fashion.