Combining image processing and machine learning to identify invasive plants in high-resolution images

This study investigates the combination of image processing and supervised classification to identify invasive yellow flag iris (YFI; Iris pseudacorus) plants in images collected by an un-calibrated, visible-light camera carried aloft by an unmanned aerial vehicle. Specifically, the image-processing steps of colour thresholding, template matching, and/or de-speckling prior to training a supervised random forest classifier are explored in terms of their benefits towards improving the resulting classification of YFI plants within an image. The impacts of performing feature selection prior to training the random forest classifier are also explored. This analysis demonstrates the importance of image processing when preparing images for classification and reveals that applying the image-processing steps of colour thresholding and de-speckling prior to classification by a random forest classifier trained to identify patches of YFI plants using spectral and textural features provided the best results.