Francesca Palumbo

This paper presents a comparative study of a design flow for generating Convolutional Neural Network (CNN) accelerators on Field Programmable Gate Arrays (FPGAs), based on an extension of the Multi-Dataflow Composer (MDC) tool, against established frameworks: HLS4ML, FINN and Vitis AI. The proposed design flow explores a previously untapped area of the design space: runtime reconfigurable accelerators. By enabling runtime reconfigurability, it provides adaptivity support, filling a gap in current FPGA-based accelerator design options. The analysis focuses on the trade-offs and benefits of each approach, particularly regarding performance and adaptivity.

Manufacturing processes increasingly depend on advanced production machinery that must deliver high quality and large volumes. This applies to die-bonding machines as well that, especially at the time being after the years of shortage, need to meet very high standards of speed and accuracy. To achieve this, these devices are exploring the use of computer vision algorithms for automatic recognition of wafer positioning and die size. Nevertheless, these systems are typically managed by software-only solutions, which may fall short under stringent execution time requirements. A promising solution is the use of heterogeneous platforms, combining general-purpose processors with reconfigurable hardware. Such platforms offer the flexibility to handle both software tasks, which benefit from operating system support, and critical functions requiring hardware acceleration. This paper presents a closed-loop implementation of a vision-based multi-sensor control system for an industrial application. The implementation exploits the capabilities of System on Module (SoM) technologies to provide flexible I/O and software execution coupled with computing acceleration for the vision algorithm on the reconfigurable Field Programmable Gate Array (FPGA) fabric. The FPGA coprocessor has been designed leveraging the High-Level Synthesis (HLS) technology and optimized on a dataset of 10k realistic images to meet the industrial use case's performance, communication, and accuracy requirements. Moreover, the resulting accelerator performance and resource utilization demonstrate the possibility of reaching state-of-the-art metrics of handwritten hardware designs while allowing for higher abstraction and productivity of the design process.

The MYRTUS Horizon Europe Project embraces the principles of the EUCloudEdgelOT Initiative and integrates edge, fog and cloud computing platforms, leveraging a cognitive engine based on swarm intelligence and federated learning to orchestrate collaborative distributed and decentralised components.

Image and video processing are one of the main driving application fields for the latest technology advancement of computing platforms, especially considering the adoption of neural networks for classification purposes. With the advent of Cyber Physical Systems, the design of devices for efficiently executing such applications became more challenging, due to the increase of the requirements to be considered, of the functionalities to be supported, as well as to the demand for adaptivity and connectivity. Heterogeneous computing and design automation are then turning into essential. The former guarantees a variegated set of features under strict constraints (e.g., by adopting hardware acceleration), and the latter limits development time and cost (e.g., by exploiting model-based design). In this context, the literature is still lacking adequate tooling for the design and management of neural network hardware accelerators, which can be adaptable and customizable at runtime according to the user needs. In this work, a novel almost automated toolchain based on the Open Neural Network eXchange format is presented, allowing the user to shape adaptivity right on the network model and to deploy it on a runtime reconfigurable accelerator. As a proof of concept, a Convolutional Neural Network for human/animal classification is adopted to derive a Field Programmable Gate Array accelerator capable of trading execution time for power by changing the resources involved in the computation. The resulting accelerator, when necessary, can consume 30% less power on each layer, taking about overall 8% more time to classify an image.

An early estimation of the exact number of fruits, flowers, and trees helps farmers to make better decisions on cultivation practices, plant disease prevention, and the size of harvest labor force. The current practice of yield estimation based on manual counting of fruits or flowers by workers is a time consuming and expensive process and it is not feasible for large fields. Automatic yield estimation based on robotic agriculture provides a viable solution in this regard. In a typical image classification process, the task is not only to specify the presence or absence of a given object on a specific location, while counting how many objects are present in the scene. The success of these tasks largely depends on the availability of a large amount of training samples. This paper presents a detector of bunches of one fruit, grape, based on a deep convolutional neural network trained to detect vine bunches directly on the field. Experimental results show a 91% mean Average Precision.