I’m Mattia Balestra, an Italian PhD student at the beginning of his second year. I have a master’s degree in “Forest Science” and I’m part of the SFARM research centre. I’m working in different project of the SFARM. One of them is related to the integration of geomatic techniques for the 3D representation of a monumental tree. Moreover, I also used remote sensing technologies to classify a forest area trough the time-series outliers’ removal. Other projects and activities are currently in progress and they involve the landscape analysis.
The “SFARM” is a Smart Farming Research and Service Centre, and it is a scientific structure of “Università Politecnica delle Marche” (Italy) which allows synergies and collaborations among different expertise. It promotes knowledge integration by means of multidisciplinary activities, the enhancement of laboratory equipment to boost research activities and academic and professional training in Precision/Smart Farming with the involvement of companies and institutions working in the local community. There are three departments that are part of the SFARM: The Department of Agricultural, Food and Environmental Sciences, the Department of Information Engineering and the Department of Construction, Civil Engineering and Architecture. This centre leads several research activities in the context of advanced digital technologies for sustainable agricultural production (Fig. 1). The “AI4SmallScaleFarming” is one of them. In this project a stabilized gimbal was used to acquire data in two fully irrigated crops, using two multi-spectral cameras operating in the NIR and in the VIS ranges. Then, a team of agronomist labelled the images to build the dataset, catching weeds by their shape and their spectral info. By this work, Artificial intelligence can support machines removing weeds by mechanical action. Same technology was used to estimate the number of olive fruits and their ripening stage. A detection network was used to isolate the olives and a bounding box was obtained for each olive. The background was isolate by means of different approaches such as the R-CNN and the Ellipse detector and the ripening estimation was carried out using classical machine learning approaches. There are still some challenging issues, such as the unbalancing of the dataset and the shadows over the olives. Another project involved monitoring systems to enhance piedmont farming. The data were collected in a chestnut and a hazelnut grove by means of weather stations and different ground sensors mounted at different ground depth and, moreover, by imagery collected using drones and satellites. All these data were analyzed as time-series to evaluate the groves’ healthy status. The BioCereals 4.0 project, on the other hand, uses Sentinel-2 and PlanetScope satellite’s images to analyze wheat crop production. Using the FPCA on vegetational index values, such as the NDVI, the values were analyzed by their principal component and the time-series of each pixel was reconstructed and isolated using them. By the FPCA-Kmeans, three cluster have been individualized. At the end, a yield map was realized for the same field in the same time span to evaluate the difference in term of confidence of the predictive models built. The last project involved a mobile laser scanner named Kaarta Stencil-2. Its LiDAR unit was used to measure olive tree crown volumes under different pruning techniques. With this Laser scanner we collected a point cloud of an olive grove pre- and post-pruning and we extract different tree metric data such as the height and the diameter at breast height of each tree. The first thanks to the normalization of the above ground point and the second thought the vertex triangulation algorithms. Then, by manual filtering in CloudCompare software, it is possible to remove the crown, and through the QSM algorithm, a 3D model is reconstructed in the 3DForest software. Thanks to these models, we were able to extract the wood volume of each tree. By the reconstruction of the olive tree crown by different algorithms, we calculated the crown volume and we compared the results with the ground truth collected in the field, in order to set the best algorithms for the olive tree volume calculation.
Mattia Balestra1, Adriano Mancini2, Roberto Pierdicca3
1 Università Politecnica delle Marche, Dipartimento di Scienze Agricole, Alimentari e Ambientali, 60131 Ancona, Italy, m.balestra@pm.univpm.it
2Università Politecnica delle Marche, Dipartimento di Dipartimento di Ingegneria dell’Informazione, 60131 Ancona, Italy, a.mancini@staff.univpm.it
3 Università Politecnica delle Marche, Dipartimento di Dipartimento di Ingegneria Civile, Edile e dell’Architettura, 60131 Ancona, Italy, r.pierdicca@staff.univpm.it