Stan Anderson K. Holaysan, Irie Aldana, Christian V. Maderazo
Plants are vital to any ecosystem and are good sources of food and oxygen. An essential condition for plants to grow well is adequate amounts of light. The aim of the study was to create a light augmentation system that would monitor light conditions of lettuce plants based on their leaves and provide artificial lighting for the plants that lack light. Image data to train the neural network used in the study was collected from two setups containing lettuce plants, one with more amounts of sunlight, and another with less. The leaves in the collected images were annotated and labelled based on the setup the image belonged to. A Mask RCNN was trained using the images to detect light conditions of lettuce leaves, while a smartphone, smart switches and artificial lights were used in the experimental light augmentation system to take images of the plants and control the artificial lights as needed. The trained Mask RCNN had poor statistical metrics, having a mean Average Precision of 0.3092, a mean Average Recall of 0.1754, and an Fl-Score of 0.2238. However, when manually tested on the experimental light augmentation system, the neural network performed accurately in labelling light conditions, although there could be improvements in accuracy scores, leaf detection and generalization to wider varieties of images. Plant growth in the experimental light augmentation system was not ideal. However, it could be observed that the plants grew better than those in the image collection setup with less sunlight, but worse than those in the setup with more sunlight. © 2022 IEEE.
University of San Carlos, Department of Computer, Information Sciences, and Mathematics, Cebu City, Philippines
