Optimised Detection of Anredera Cordifolia (Madeira Vine) using a Mask-RCNN and Anredera Cordifolia’s prominent features as object classes

Abstract

The research interest in plants using current technologies is growing, and with it is plant recognition using deep Convolutional Neural Networks (CNNs). The CNNs and its variants such as RCNN have recently become a popular method of plant feature recognition due to their superior ability to classify, detect, and label features with high fidelity. Anredera cordifolia, also known as Madeira Vine, is a plant species that unnecessarily invade environments, hence destroying the plants occupying those environments. Here, we develop a computer vision model for the detection of Anredera cordifolia with Mask-RCNN for use in environments that may need drones to detect the presence of this foreign plant species. To optimize the model’s confidence in detecting the presence of the Anredera cordifolia, a Mask-RCNN was trained with images of the Anredera cordifolia using three distinct features of this plant as object classes. These features that were used to build classes of the Mask-RCNN were leaves, flowers, and the tubers. This novel approach ensures the detection of the Anredera cordifolia as the prominent features were used as class objects of the Mask-RCNN. The results of the experiments showed that the Mask- RCNN was able to overlay masks and bounding boxes around the Anredera cordifolia features that were detected.