SHD-O2D: A Novel Image Dataset for Super-High-Density Olive Orchard Yield Estimation

Extra virgin olive oil is probably the most representative agricultural product of the Mediterranean area. Mechanization and automation in olive production processes have achieved considerable technological advances, especially in super-high-density orchards. Nowadays, further progress has been marked by the integration of modern technologies based on Artificial Intelligence, which are rapidly gaining traction also within the agricultural sector. Specifically, Computer Vision systems are emerging as valuable tools for several agricultural tasks, including crop monitoring, plant disease detection, and orchard yield estimation. This last fact increasingly points to the need for specific image datasets, which are essential for training these models. In line with these considerations, we present in this work a novel super-high-density olive orchard image and data set, as well as its first application in training a Computer Vision system based on a Weakly Supervised paradigm. The dataset, named $SHD-O^{2}D$ (Super-High-Density Olive Orchard Dataset) was collected from a single row of Piantone di Mogliano olive cultivar trees, with 8 images captured per plant under different conditions: sides of the tree, weather conditions, and the presence or absence of fruits. The results achieved demonstrate the effectiveness of applying Computer Vision systems for yield estimation in olive orchards.