Recent studies have shown evidence of a significant decline of the Posidonia oceanica meadows on a global scale. The monitoring and mapping of these meadows and its marine habitat are fundamental tools for measuring its status and growth opportunities. The presence of hard substrates benefits P. oceanica survival and development rates. We present an approach based on a deep neural network to automatically perform a high-precision semantic segmentation of Posidonia oceanica meadows and its seafloor habitat in sea-floor images, offering several improvements over the state of the art techniques. The presented network is able to accurately distinguish the most relevant classes: P. oceanica meadows, and rocky and sandy areas.
Publication type: Conferences
Machine learning techniques are attracting a huge amount of interest from both industry and academia. For instance, Convolutional deep Neural Networks (CNNs) have recently enjoyed a notable success in image understanding. The automotive industry is already using image classifiers for Advanced Driver-Assistance Systems and in the development of the upcoming autonomous cars, which will have to guarantee high levels of reliability. The certification of systems based on machine learning is an open issue but it is clear that any improvement in the performance of image classifiers is to be welcomed. CNNs need to be trained to act as image classifiers. This training leads to slightly different classification capacity depending on some training parameters. In this paper we present a first exploration on the use of schemes based on voting on the results of several CNNs trained differently, as a means to increase the final classification performance, and thus the reliability, of this type of systems.
Critical Adaptive Distributed Embedded Systems (ADES) must meet high real-time and dependability requirements, while autonomously rearranging themselves to operate in dynamic operational contexts. The DFT4FTT project proposes a self-reconfigurable complete infrastructure, whose different architectural levels provide a set of real-time (RT), fault-tolerance (FT) and flexibility mechanisms that collaborate to adequately support critical ADESs. To efficiently tolerate transient faults in the network of an ADES, this paper describes our ongoing work on providing a dynamic temporal replication of messages that takes into account all the DFT4FTT fault-tolerance mechanisms from a holistic point of view.
Critical Adaptive Distributed Embedded Systems (ADESs) are nowadays the focus of many researchers. ADESs are envisioned to dynamically modify their behavior to support changes of their real-time and dependability requirements at runtime as the conditions of the environment in which they operate vary. To provide ADESs with an adequate communication infrastructure, our research group proposed the Flexible-Time-Triggered Replicated Star (FTTRS). FTTRS provides highly reliable communication services on top of Ethernet, while keeping the adaptivity benefits that the Flexible- Time-Triggered (FTT) communication paradigm offers from a real-time perspective. This paper formally verifies, by means of model checking, the correctness of the mechanisms FTTRS includes to enforce consistent changes of the communication scheduling at runtime.
The Audio Video Bridging Task Group (AVB TG) from the IEEE proposed a series of standards to provide Ethernet with soft real-time guarantees. Later on, the group was renamed to Time-Sensitive Networking and its scope was broadened to provide new services to support critical applications. The Stream Reservation Protocol (SRP) stands out among the projects developed by the groups. Nonetheless, SRP was originally designed for audio/video applications and does not take into account properties that are important for critical systems; such as termination and consistency. In this work we study the termination and consistency of SRP at different levels, using a model we developed of this protocol in UPPAAL . We see that SRP does not provide termination nor consistency, we discuss how this can impact critical applications and we propose solutions for all the issues detected.
The Time-Sensitive Networking (TSN) Task Group (TG) is providing Ethernet with timing guarantees, reconfiguration services and fault tolerance mechanisms. Some of TSN’s targeted applications are real-time critical applications, which must provide a correct service continuously. To support these applications the TSN TG standardised a spatial redundancy mechanism. Even though spatial redundancy can tolerate permanent and temporary faults, it is not cost-effective. Instead, temporary faults can be tolerated using time redundancy. We proposed the Proactive Transmission of Replicated Frames (PTRF) mechanism to tolerate temporary faults in the links. In this work we present a new PTRF approach, a PTRF simulation model and a comparison of the approaches using exhaustive fault injection.
Distributed Embedded Systems (DES) typically have real-time and dependability requirements. Moreover, if they have to operate in dynamic operational contexts, they need to be adaptive. That is, they must be able to automatically and autonomously rearrange in response to changes. In order for a DES to be adaptive, its underlying subsystems must be flexible. The implementation of the flexibility, just like the implementation of the real-time and dependability, cannot be done in an orthogonal manner since it entails the collaboration of various subcomponents at different levels of the architecture. The DFT4FTT project proposes a self-reconfigurable infrastructure for implementing DES with real-time, reliability and adaptivity requirements. One of the most relevant fault tolerance mechanisms is the dynamic replication of messages, that makes it possible to tolerate transient faults affecting the network. In this paper we describe more in-detail the design and implementation of this mechanism.
The Audio Video Bridging Task Group (AVB TG) from the IEEE proposed a series of standards to provide Ethernet with soft real-time guarantees. Later on, the group was renamed to Time-Sensitive Networking and its scope was broadened to provide new services to more critical applications. The Stream Reservation Protocol (SRP) stands out among the projects developed by the groups. Nonetheless, SRP was originally designed for audio/video applications and does not take into account properties that are important for critical systems; such as termination and consistency. In this work we study the termination and consistency of the reservations using AVB’s SRP. We used Uppaal to model the protocol and to verify the properties. We see that SRP does not provide these properties, we discuss how this can impact critical applications and we propose solutions for some of the issues detected and a roadmap to solve the rest.
The Time Sensitive Networking (TSN) Task Group has been working on describing a set of standards that will provide enhanced capabilities to standard Ethernet. Specifically, they work to provide Ethernet with real-time, reconfiguration and reliability capacities. Nevertheless, this set of standards (commonly referred to as TSN) does not cover some reliability aspects that are relevant for the correct operation of critical distributed control systems. Thus, in this work we present a first proposal of a highly reliable architecture and a set of mechanisms based on TSN to support the real-time and reliability requirements of these critical systems.