“How can we use deep learning techniques to measure flower richness and density in agricultural areas?”
Flower Power is a research group composed of people from different organizations, such as the Naturalis Biodiversity Center, JADS Research, and the Fontys Research Group “AI & Big Data”. They work with big research projects related to biodiversity and sustainability. Agricultural intensification has resulted in the loss of biodiversity and ecosystem on-farm areas. The main negative impact of intensive use of pesticides, herbicides, and fertilizers has caused the reduction of the availability of flowering plants, thus reducing nutritional resources for insects and a huge decline in species, such as bees and other pollinators. To tackle the issue of the decline of wildflowers, Flower Power is looking for a solution that will help recognize and count species of flowers from drone images by predicting, evaluating, and giving strong insights for future improvements. For this, they want a deep learning algorithm that satisfies these conditions.
Our team’s goal was to do the ground research and experiment with deep learning models to see what worked. We were aware that it was likely that we would not have a prototype at the end. The dataset consisted of images taken with drones and a smaller subset of pictures taken with phones over field areas. The observations were in an excel file we received. The data received was not directly labeled, we did not know where the flower individuals were actually located in the pictures. Even with the high resolution of the pictures, the flowers themselves when cropped were very blurry.