A Smart Fault Detection Scheme for Reliable Image Processing Applications

Abstract

Traditional fault detection/tolerance techniques exploit multiple instances of the nominal processing and then perform a bit-wise comparison of the outputs to detect the occurrence of faults. In specific application scenarios, e.g., image/signal processing, the elaboration has an inherent degree of fault tolerance because it is possible to use the output even in the presence of slight alterations. In these contexts, the classical bit-wise comparison may be inefficient. Indeed, it may lead to conservatively discard outputs that have been only slightly altered by the fault and that could still be usefully exploited. In this paper, we propose a smart checking scheme based on Convolutional Neural Networks that rather than distinguishing between faulty and not faulty images, discriminates between usable and not usable images according to the ability of the end user to correctly process the output. The experimental evaluation shows that this solution enables an execution time saving of about 6.35% with a 99.42% accuracy, on average.

An Approximation-based Fault Detection Scheme for Image Processing Applications

“…the traditional DWC approach and from 22% to 34% w.r.t. the proposal in [4], with a false negative rate (images that are not usable by the end-user application that are wrongly not discarded) always lower than 1%, that is acceptably low and comparable with the one in [4].…”

“…Let us consider as an example a satellite that takes aerial pictures of the Earth, executes a pipeline of filters on the captured images and sends then the output images to a ground station. As it has already been discussed by Biasielli et al [4], if a fault affects the execution of the image processing filters, it would be highly beneficial for the satellite to be able to decide whether the corrupted output image is unusable by the application running on the ground station or it is still usable although being corrupted. In the latter case, re-execution of the application would be avoided, thus saving time.…”

“…We propose a lightweight fault detection scheme for image processing applications that combines image usability clas-sification with approximate computing to reduce the time overhead introduced by duplication and to avoid unnecessary re-executions, thus, further saving time. We borrow from [4] the idea of classifying output images based on their usability by the final end-user application and the use of a Convolutional Neural Network (CNN) as a checker instead of the classical Two-Rail Checker (TRC). On the other hand, we extend the proposal in [4] by introducing approximate computing.…”

“…We borrow from [4] the idea of classifying output images based on their usability by the final end-user application and the use of a Convolutional Neural Network (CNN) as a checker instead of the classical Two-Rail Checker (TRC). On the other hand, we extend the proposal in [4] by introducing approximate computing. Indeed, instead of having two exact replicas of the pipeline, our hardening scheme pairs the nominal pipeline producing output images at the minimum quality level required by the application designer with a redundant approximated counterpart.…”

“…The defined fault detection scheme allows for a significant time saving w.r.t. both the classical DWC and the proposal in [4] without loosing fault detection capability.…”