Université de technologie de Troyes open archive
Not a member yet
10722 research outputs found
Sort by
A sustainable Two-Echelon waste collection routing: Understanding waste picker behavior
No full textInternational audienceWaste collection is a key component of reverse logistics that represents a significant challenge for low- and middle-income countries, where the process is highly labor-intensive and often informal. The informal sector plays an important role in solid waste collection, however its operational reality is rarely considered in routing models. This paper aims to analyze how the behavior of waste pickers influences route definition by considering sustainability factors and the impact of node characteristics, the use of a heterogeneous fleet, stochastic demand, and waste price. The problem is modeled as a Two-Echelon Waste Collection Vehicle Routing Problem (2E-WCVRP) integrating the three components of sustainability, economic, environmental, and social. Social aspects and informality in routing, often overlooked in literature, are key considerations. The proposed model considers interactions between waste pickers (first echelon) and trucks (second echelon), incorporating factors such as profit, workload distribution, CO emissions, and social impacts. The waste pickers unload waste bags at satellite nodes, where urban trucks later collect them. The objective is to maximize the profit of the waste pickers and minimize the total costs of the urban trucks. A two-stage solution approach is developed, employing a metaheuristic based on Ant Colony Optimization (ACO). A key innovation is the integration of waste pickers’ decision-making behavior into the pheromone update mechanism, considering factors like profit maximization and node-vehicle compatibility. The model and solution methodology are demonstrated through test instances and a case study in Cali, Colombia. Experimental results demonstrate the effectiveness of this approach in optimizing routing and improving sustainability
Single-Layer Distillation with Fourier Convolutions for Texture Anomaly Detection
No full textInternational audienceIn industrial quality control, detecting anomalies in visual textures is essential for ensuring product quality and operational efficiency. Early identification of defects prevents faulty items from reaching consumers, reduces waste, and maintains high standards of production. Numerous unsupervised anomaly detection methods heavily depend on the integration of multiple layers from various pretrained models, a selection often made through empirical means. We propose SingleNet, an innovative knowledge distillation approach tailored for fast unsupervised texture anomaly detection, using a single layer from a compact pretrained model. Contrary to the previous knowledge distillation approaches, our network leverages fast Fourier convolutions (FFC) to reconstruct a degraded version of the teacher extracted features. At test time, we employed a frequency-aware filtering mechanism to reduce reconstruction artifacts caused by discrepancies between teacher and student architectures. Empirical results demonstrate the efficacy of our approach, attaining state-of-the-art performance across evaluated datasets coupled with expedited high-speed inference
Hybrid AI Pipeline for Laboratory Detection of Internal Potato Defects Using 2D RGB Imaging
No full textThe internal quality assessment of potato tubers is a crucial task in agro-industrial processing. Traditional methods struggle to detect internal defects such as hollow heart, internal bruises, and insect galleries using only surface features. We present a novel, fully modular hybrid AI architecture designed for defect detection using RGB images of potato slices, suitable for integration in industrial sorting lines. Our pipeline combines high-recall multi-threshold YOLO detection, contextual patch validation using ResNet, precise segmentation via the Segment Anything Model (SAM), and skin-contact analysis using VGG16 with a Random Forest classifier. Experimental results on a labeled dataset of over 6000 annotated instances show a recall above 90\% and precision near 100\% for most defect classes. The approach offers both robustness and interpretability, outperforming previous methods that rely on costly hyperspectral or MRI techniques. This system is scalable, explainable, and compatible with existing 2D imaging hardware
Explainable AI Reveals Hidden Biomarkers in Parkinson's Brain Imaging
No full textInternational audienceBackground As new SARS-CoV-2 variants emerge and as treatment of COVID-19 ARDS remains exclusively supportive, there is an unmet need to better characterize its different phenotypes to tailor personalized treatments. Clinical, biological, spirometric and CT data hardly allow deciphering of Heavy (H), Intermediate (I) and Light (L) phenotypes of COVID-19 ARDS and the implementation of tailored specific strategies (prone positioning, PEEP settings, recruitment maneuvers). We hypothesized that the ratio of two pivotal COVID-19 biomarkers (interleukin 6 [IL-6] and Krebs von den Lungen 6 [KL-6], related to inflammation and pneumocyte repair, respectively) would provide a biologic insight into the disease timeline allowing 1) to differentiate H, I and L phenotypes, 2) to predict outcome and 3) to reflect some of CT findings. Methods and findings This was a retrospective analysis of prospectively acquired data (COVID HUS cohort). Inclusion concerned any patient with severe COVID-19 pneumonia admitted to two intensive care units between March 1 st and May 1 st , 2020, in a high-density cluster of the first epidemic wave (Strasbourg University Hospital, France). Demographic, clinical, biological (standard, IL-6 [new generation ELISA], KL-6 [CLEIA technique]), spirometric (driving pressure, respiratory system compliance) and CT data were collected longitudinally. CT analysis included semi-automatic and automatic lung measurements and allowed segmentation of lung volumes into 4 (poorly aerated, non-aerated, overinflated and normally aerated) and 3 (ground-glass, restricted normally aerated, and overinflated) zones, respectively. The primary outcome was to challenge the IL-6/KL-6 ratio capacity to decipher the three COVID-19 ARDS phenotypes (H, I and L) defined on clinical, spirometric and radiologic grounds. Secondary outcomes were the analysis of the prognostic value of the IL-6/KL-6 ratio and its correlates with CT-acquired data. Multivariate analysis was based on principal component analysis. One hundred and forty-eight ventilated COVID-19 ICU patients from the COVID HUS cohort were assessed for eligibility and 77 were included in the full analysis. Most were male, all were under invasive mechanical ventilation and vasopressor therapy and displayed high severity scores (SAPSII: 48 [42–56]; SOFA: 8 [7–10]). The L, I and H COVID ARDS phenotypes were identified in 11, 15 and 48 patients, respectively. In three patients, the phenotype could not be defined precisely. Thirty patients (39%) died in the ICU and the number of ventilator-free days was 2 [0–2] days. The IL-6/KL-6 ratio was not significantly different between the L, I and H phenotypes and evolved according to similar patterns over time. Surviving and deceased patients displayed an inverse kinetic of KL-6. IL-6 and the IL-6/KL-6 ratio were linearly associated with ground-glass volume on semi-automatic and automatic CT lung measurements. Conclusions In our population of severe ventilated COVID ARDS patients, the IL-6/KL-6 ratio was not clue to differentiate the H, I and L phenotypes and tailor a personalized ventilatory approach. There was an interesting correlation between IL-6/KL-6 ratio and ground-glass volume as determined by automated lung CT analysis. Such correlation deserves more in-depth pathophysiological study, at best gathered from a prospective cohort with a larger sample size and histological analysis. Trial registration COVID HUS Trial registration number: NCT0440572
Beyond the Striatum: A Whole-Brain Approach for Parkinson's Disease Diagnosis Using 3D CNNs and DaTSCAN SPECT Imaging
No full textInternational audienceParkinson's disease (PD) is a progressive neurodegenerative disorder that primarily affects motor function due to dopamine loss in the substantia nigra. While DaTSCAN SPECT imaging is nowadays widely used for diagnosis confirmation, conventional methods focus on the striatum, potentially overlooking significant biomarkers present in other brain regions. This study explores the use of 3D Convolutional Neural Networks (3D CNNs) to classify PD stages using both full-brain and cropped DaTSCAN volumes. The model was trained using 10-fold cross-validation with augmentation involving translation and rotation augmentations and evaluated based on many evaluation metrics. The results indicate that both full-sized and cropped datasets achieved high classification performance. The full-sized dataset exhibited clearer stage-specific differences, however the cropped dataset showed greater overlap in features. Kernel Density Estimation (KDE) analysis further confirmed that neurodegenerative changes extend beyond the striatum, reinforcing the importance of whole-brain analysis. These findings suggest that restricting Artificial Intelligence (AI) models to predefined regions may limit the identification of key imaging biomarkers for PD progression. Future research should focus on multi-modal imaging integration and explainable AI techniques to enhance early detection and improve clinical decision-making
Brownian Feature Trajectories with Geometric Decision Boundaries for Edge Anomaly Detection
No full textEmbedded sensor systems deployed in heterogeneous and evolving environments face strong constraints in power, supervision, and connectivity, which limit the applicability of classical anomaly detection methods. This paper introduces a lightweight framework for unsupervised anomaly detection based on Brownian feature trajectories. Each sensor event is modeled as a stochastic trajectory whose drift and covariance parameters summarize the nominal temporal evolution of features. To overcome the limitations of purely parametric models, the framework integrates a geometric decision layer that combines kernel density estimation with convex envelopes, enabling robust detection of trajectories that traverse low-density or out-of bound regions. The selection of the kernel bandwidth and decision thresholds is guided by data-driven statistical rules to ensure interpretability and consistent sensitivity across devices. A statistical consistency test, based on Gaussianity and covariance conformity of increments, provides additional validation of deviations from the Brownian hypothesis. The proposed approach achieves interpretable, adaptive, and computationally efficient anomaly detection suitable for real-time edge deployment in isolated infrastructure monitoring and surveillance scenarios, where data remain specific to each use case
Answering Application of Generative AI in Industry: Integration of Semantic Representation
No full textInternational audienceGenerative AI acts as an effective guide in decision-making process. This paper is designed to outline the main challenges in applying this technique within businesses and for specific activities. Semantic representation as ontology can be one solution for these challenges. Our first work to link ontology to LLM algorithms is illustrated in financial institution
De quoi El Risitas est-il le mème ? Analyse sémiopragmatique
No full textInternational audienceThis study focuses on the use of the El Risitas meme in the Blabla18-25 forum and addresses three aspects. We first proceed to a semiopragmatic analysis of the meme. By analyzing a corpus of messages extracted from the Blabla18-25 forum, from the jeuxvideo.com website, including El Risitas in its original or transformed forms, we take into account the following different aspects: forms (visual or scripto-visual language), type of propagation (replication or variation) and functions of the meme. This analysis highlights the polysemic nature of the meme and its role in the development of a community language. We then propose a dialogical analysis to describe the function of the meme in the interactional dynamics of exchanges. Finally, we show how the meme has become a political object, enabling its users to express their proximity to a set of extreme-right, racist, anti-feminist or homophobic opinions
Models and tools for supporting sustainability assessment in Systems Engineering
No full textInternational audienceSince several years, sustainability has become a very important challenge for our societies. Our lifestyles are in the process of making our planet uninhabitable because of the various impacts that we, as human beings, are inflicting on it. As part of these impacts, we focus on Complex Systems designed by humans. It is of uttermost importance to be able to analyze and design complex systems so that the sustainability features are taken into account. More specifically, the contribution of this article is to propose models and tools for the assessment of systems sustainability during the analysis and design phases. The analysis and design of systems may be difficult tasks. It is even more so for complex systems. Since decades, the System Engineering (SE) field has given birth to a family of systemic and multidisciplinary approaches for the design of systems. Among SE approaches, Model-Based System Engineering (MBSE) is a special kind of SE that relies on formalized models as first-class citizens deliverables for all analysis and design activities from requirements elicitation to final design and validation. SysML (Systems Modelling Language) is one of these MBSE approaches. SysML is a well-known, general purpose graphical systems modelling language that supports SE approaches. In addition, SysML allows its own language extension by the creation of new concepts and diagrams. This extension mechanism is known as Domain Specific Modelling Language (DSML). The contributions presented in this paper consist in the definition of an extension of SysML that provides models and tools in order to assess the sustainability of systems during analysis and design. This extension of SysML is based upon a technique, named profile, and proposes new modelling concepts for taking into consideration sustainability issues during systems engineering. Moreover, these new model elements are supported by software tools issued from the MBSE domain and allow the development of ad-hoc software tooling support
Gold nanoparticles combined with ultrafine TiO 2 layer: a reliable probe for Raman thermometry
No full textInternational audienceTemperature determination methods in metal nanoparticles are essential for providing information on energy dissipation dynamics in such systems and for temperature-sensitive applications, hence the need for high-performance thermometry techniques is evident. In this study, we propose new efficient probes for Raman antiStokes-Stokes thermometry based on gold nanoparticles (AuNPs), prepared by thermal dewetting and controllably functionalized with a 2 nm TiO2 layer by the sol–gel method. AuNPs@TiO2 demonstrated good stability and a usable response over a temperature range from 25 °C to 240 °C generated by external thermal and thermoplasmonic heating of the sample. We validate the methodology by taking into account the spectral efficiency of the Raman spectrometer as well as the extinction properties of AuNPs@TiO2 in the calculations. This approach enables us to propose a reliable temperature measurement over two hundred degrees range