Genetic Algorithm Feature Selection and Classifier Optimization Using Moment Invariants and Shape Features

In row spot spraying technology has been in the constant development as a means to reduce the application of herbicide. This will enable higher efficiency in chemical usage and reduce impact on environment. A vision system was develop for a spot spraying mechanism on an autonomous ground vehicle. The aim is to enable autonomous spot spraying of selective herbicide crop rows. However the classifier is required to be fine tune. In this research, an autonomous fine tuning and feature selection using Genetic algorithm (GA) was proposed and tested with the assumption that the weeds are young and non-occluded. The results show a feasible means to distinguished the weeds using the selected features (which are a combination of fractal, shape features and HU moment invariants. The results show that solidity of the shapes are the most prominent feature and alone could be used to achieved 90% recognition rates. 100% Recognition was achieved with the combination of shape and moment invariants.