Evaluation of a 2D multileader training system for improving sustainability and precision orchard management applications in italian apple orchards

The presented study aimed to evaluate the productive and physiological behavior of a 2D multileader apple training systems in the Italian environment both investigating the possibility to increase yield and precision crop load management resolution. Another objective was to find valuable thinning thresholds guaranteeing high yields and matching fruit market requirements. The thesis consists in three studies carried out in a Pink Lady®- Rosy Glow apple orchard trained as a planar multileader training system (double guyot). Fruiting leaders (uprights) dimension, crop load, fruit quality, flower and physiological (leaf gas exchanges and fruit growth rate) data were collected and analysed. The obtained results found that uprights present dependence among each other and as well as a mutual support during fruit development. However, individual upright fruit load and upright’s fruit load distribution on the tree (~ plant crop load) seems to define both upright independence from the other, and single upright crop load effects on the final fruit quality production. Correlations between fruit load and harvest fruit size were found and thanks to that valuable thinning thresholds, based on different vegetative parameters, were obtained. Moreover, it comes out that an upright’s fruit load random distribution presents a widening of those thinning thresholds, keeping un-altered fruit quality. For this reason, uprights resulted a partially physiologically-dependent plant unit. Therefore, if considered and managed as independent, then no major problems on final fruit quality and production occurred. This partly confirmed the possibility to shift crop load management to single upright. The finding of the presented studies together with the benefits coming from multileader planar training systems suggest a high potentiality of the 2D multileader training systems to increase apple production sustainability and profitability for Italian apple orchard, while easing the advent of automation in fruit production.

Silence after bombs. Practice, role and training of interpreters in conflict zones. A case study of the wars in Bosnia and Herzegovina and Croatia.

This dissertation investigates the role, training and practice of the interpreters that worked during the wars in Bosnia and Herzegovina and Croatia in the 1990s, both at a high political level and on the ground for peackeeping troops. Adopting a historical method that uses interviews, newspaper articles, videos, archival documents and pictures the author tries to retrace how those interpreters were hired, employed and what challenges they faced in their daily work. The aim is to give voice to a category that has long been forgotten, to investigate how mediated interaction is shaped by violent conflict and to offer hindsight to improve the recruitment and management of local interpreters by armed forces.

Directly training spiking neural networks for cyber-physical systems: from supervised to reinforcement learning

Spiking Neural Networks (SNNs) are bio-inspired Artificial Neural Networks (ANNs) utilizing discrete spiking signals, akin to neuron communication in the brain, making them ideal for real-time and energy-efficient Cyber-Physical Systems (CPSs). This thesis explores their potential in Structural Health Monitoring (SHM), leveraging low-cost MEMS accelerometers for early damage detection in motorway bridges. The study focuses on Long Short-Term SNNs (LSNNs), although their complex learning processes pose challenges. Comparing LSNNs with other ANN models and training algorithms for SHM, findings indicate LSNNs' effectiveness in damage identification, comparable to ANNs trained using traditional methods. Additionally, an optimized embedded LSNN implementation demonstrates a 54% reduction in execution time, but with longer pre-processing due to spike-based encoding. Furthermore, SNNs are applied in UAV obstacle avoidance, trained directly using a Reinforcement Learning (RL) algorithm with event-based input from a Dynamic Vision Sensor (DVS). Performance evaluation against Convolutional Neural Networks (CNNs) highlights SNNs' superior energy efficiency, showing a 6x decrease in energy consumption. The study also investigates embedded SNN implementations' latency and throughput in real-world deployments, emphasizing their potential for energy-efficient monitoring systems. This research contributes to advancing SHM and UAV obstacle avoidance through SNNs' efficient information processing and decision-making capabilities within CPS domains.