Search

Within Artificial Intelligence, Deep Learning (DL) represents a paradigm that has been showing unprecedented performance in image and audio processing by supporting or even replacing humans in defect and anomaly detection. The Railway sector is expected to benefit from DL applications, especially in predictive maintenance applications, where smart audio and video sensors can be leveraged yet kept distinct from safety-critical functions. Such separation is crucial, as it allows for improving system dependability with no impact on its safety certification. This is further supported by the development of DL in other transportation domains, such as automotive and avionics, opening for knowledge transfer opportunities and highlighting the potential of such a paradigm in railways. In order to summarize the recent state-of-the-art while inquiring about future opportunities, this paper reviews DL approaches for the analysis of data generated by acoustic and visual sensors in railway maintenance applications that have been published until August 31st, 2021. In this paper, the current state of the research is investigated and evaluated using a structured and systematic method, in order to highlight promising approaches and successful applications, as well as to identify available datasets, current limitations, open issues, challenges, and recommendations about future research directions.

The growing availability of cheap and reliable fingerprint acquisition scanners is resulting in an increasing spread of Fingerprint-based Authentication Systems (FAS) in consumer electronics. This has giving rise to a new wave in research on both smarter spoofing attacks, aimed to bypass a FAS by using a counterfeit fingerprint, and on more effective Liveness Detectors (LD), aimed to discern authentic (live) fingerprints from fake ones. As in many other computer vision tasks, deep Convolutional Neural Networks (CNN) demonstrated to be very effective also for fingerprint liveness detection. However, we showed that it is possible to adapt adversarial perturbation approaches to mislead CNN-based LD. In this paper, we want to make a step further toward the design of a black-box attack by investigating whether it is possible to transfer a perturbation across different CNN liveness detectors in the case of a target LD very different from the one used to compute the perturbations. To this aim, we designed an attack scenario where a shadow LD (i.e. an adaptation of the substitute technique for the liveness detection application) is used to generate an adversarial fingerprint in a white-box setting before submitting it to the real target LD, invoked in a total back-box manner. Finally, we analysed the impact that such attack has on the authentication system, also analysing if and to what extent the scanner and the spoofing material combinations affect the success of the attack.

Lung ultrasound imaging is reaching growing interest from the scientific community. On one side, thanks to its harmlessness and high descriptive power, this kind of diagnostic imaging has been largely adopted in sensitive applications, like the diagnosis and follow-up of preterm newborns in neonatal intensive care units. On the other side, state-of-the-art image analysis and pattern recognition approaches have recently proven their ability to fully exploit the rich information contained in these data, making them attractive for the research community. In this work, we present a thorough analysis of recent deep learning networks and training strategies carried out on a vast and challenging multicenter dataset comprising 87 patients with different diseases and gestational ages. These approaches are employed to assess the lung respiratory status from ultrasound images and are evaluated against a reference marker. The conducted analysis sheds some light on this problem by showing the critical points that can mislead the training procedure and proposes some adaptations to the specific data and task. The achieved results sensibly outperform those obtained by a previous work, which is based on textural features, and narrow the gap with the visual score predicted by the human experts., Comment: 7 pages

With the increasing spread of easy and effective frameworks, in recent years Deep Learning approaches are becoming more and more used in several application fields, including computer vision (such as natural and biomedical image processing), automatic speech recognition (ASR) and time-series analysis. If, on one hand, the availability of such frameworks allows developers to use the one they feel more comfortable with, on the other, it raises questions related to the reproducibility of the designed model across different hardware and software configurations, both at training and at inference times. The reproducibility assessment is important to determine if the resulting model produces good or bad outcomes just because of luckier or blunter environmental training conditions. This is a non-trivial problem for Deep Learning based applications, not only because their training and optimization phases strongly rely on stochastic procedures, but also because of the use of some heuristic considerations (mainly speculative procedures) at training time that, although they help in reducing the required computational effort, tend to introduce non-deterministic behavior, with a direct impact on the results and on the model’s reproducibility. Usually, to face this problem, designers make use of probabilistic considerations about the distribution of data or focus their attention on very huge datasets. However, this kind of approach does not really fit some application field standards (such as medical imaging analysis with Computer-Aided Detection and Diagnosis systems – CAD) that require strong demonstrable proofs of effectiveness and repeatability of results across the population. It is our opinion that in those cases it is of crucial importance to clarify if and to what extent a Deep Learning based application is stable and repeatable as well as effective, across different environmental (hardware and software) configurations. Therefore, the aim of this work is to quantitatively analyze the reproducibility problem of Convolutional Neural Networks (CNN) based approaches for the biomedical image processing, in order to highlight the impact that a given software framework and hardware configurations might have when facing the same problem by the same means. In particular, we analyzed the problem of breast tissue segmentation in DCE-MRI by using a modified version of a 2D U-Net CNN, a very effective deep architecture for semantic segmentation, using two Deep Learning frameworks (MATLAB and TensorFlow) across different hardware configurations.

Nowadays, Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI) is increasingly succeeding as a complementary methodology for breast cancer, with Computer Aided Detection/Diagnosis (CAD) systems becoming essential technological tools to provide early detection and diagnosis of tumours. Several CADs make use of machine learning, resulting in a constant design of hand-crafted features aimed at better assisting the physician. In recent years, Deep learning (DL) approaches raised in popularity in many pattern recognition tasks thanks to their ability to learn compact hierarchical features that well fit the specific task to solve. If, on one and, this characteristic suggests to explore DL suitability for biomedical image processing, on the other, it is important to take into account the physiological inheritance of the images under analysis. With this goal in mind, in this work we propose "3TP U-Net", an U-Shaped Deep Convolutional Neural Network that exploits the well-known Three Time Points approach for the lesion segmentation task. Results show that our proposal is able to outperform not only the classical (non-deep) approaches but also some very recent deep proposal, achieving a median Dice Similarity Coefficient of 61.24%.

Supporting the early diagnosis of skin cancer is crucial for the sake of any kind of treatment or surgery. This work proposes to improve the outcome of automatic diagnoses approaches by using an ensemble of pre-trained deep convolutional neural networks and a suitable voting strategy. Moreover, a novel patching approach has been deployed. The proposal has been fairly evaluated with the literature proposals demonstrating good preliminary results.

Here we propose a system for automatic dietary monitoring of canteen customers based on robust computer vision techniques. The proposed system recognizes foods and estimates food leftovers. Results achieved on 1000 customers of a real canteen are promising.

Recognition systems based on a combination of different experts have been widely investigated in the recent past. General criteria for improving the performance of such systems are based on estimating the reliability associated with the decision of each expert, so as to suitably weight its response in the combination phase. According to the methods proposed to-date, when the expert assigns a sample to a class, the reliability of such a decision is estimated on the basis of the recognition rate obtained by the expert on the chosen class during the training phase. As a consequence, the same reliability value is associated with every decision attributing a sample to a same class, even though it seems reasonable to take into account its dependence on the quality of the specific sample. We propose a method for estimating the reliability of each single recognition act of an expert on the basis of information directly derived from its output. In this way, the reliability value of a decision is more properly estimated, thus allowing a more precise weighting during the combination phase. The definition of the reliability parameters for widely used classification paradigms is discussed, together with the combining rules employing them for weighting the expert opinions. The results obtained by combining four experts in order to recognise handwritten numerals from a standard character database are presented. Comparison with classical combining rules is also reported, and the advantages of the proposed approach outlined.

Topics extraction has become increasingly important due to its effectiveness in many tasks, including information filtering, information retrieval and organization of document collections in digital libraries. The Topic Detection consists to find the most significant topics within a document corpus. In this paper we explore the adoption of a methodology of feature reduction to underline the most significant topics within a document corpus. We used an approach based on a clustering algorithm (X-means) over the t f -id f matrix calculated starting from the corpus, by which we describe the frequency of terms, represented by the columns, that occur in each document, represented by a row. To extract the topics, we build n binary problems, where n is the numbers of clusters produced by an unsupervised clustering approach and we operate a supervised feature selection over them considering the top features as the topic descriptors. We will show the results obtained on two different corpora. Both collections are expressed in Italian: the first collection consists of documents of the University of Naples Federico II, the second one consists in a collection of medical records.