Deep learning in flower quantification of Catharanthus roseus (L.) G. Don

Abstract

Deep learning techniques are increasingly automating tasks performed manually, thanks to the robustness and precision of their results, which encourages their use as a tool in the floriculture and landscaping sector. The amount of floristic species is wide and diverse, whether in shape, texture or color. The ornamental species Catharanthus roseus, considered to have a tropical climate, it has cultivars in which the color of the flowers is one of its attractive aspects, and these can vary from white to different shades of pink. Thus, the objective of this work was to use deep learning techniques to evaluate the potential of the two-stage convolutional approach called Mask R-CNN to quantify C. roseus flowers and qualify them in terms of color for application in the floriculture and landscaping sectors. 700 images were collected in gardens in the North of Minas through smartphone cameras, of which 500 had both pink and white flowering and 200 had only the leaves, to compose the background. For the composition of the synthetic image bank, 100 white flowers and 100 pink flowers were processed in png format and formed the foreground, the two being separated as two subclasses. The training using the transfer learning technique with the Mask R-CNN algorithm was carried out in Google collaborative, with commands in python language and libraries from the Github platform. Through rating quality evaluators, the Convolutional Neural Network Mask R-CNN showed overall accuracy above 90% and accuracy above 80%. The network proved to be efficient in estimating the number of flowers, in addition to detecting and segmenting them, qualifying them in terms of color. Therefore, the methodology can be used in the floriculture and landscaping sector to estimate and quantify flowers through images.