| Titre : | Intelligent System for Plastic Detection Using Hyperspectral Imaging Techniques |
| Auteurs : | Amiri Adem, Auteur ; Laouar, Auteur ; Ahmed Tibermacine, Directeur de thèse |
| Type de document : | Mémoire magistere |
| Editeur : | Biskra [Algérie] : Faculté des Sciences Exactes et des Sciences de la Nature et de la Vie, Université Mohamed Khider, 2025 |
| Langues: | Français |
| Langues originales: | Anglais |
| Résumé : |
Plastic pollution has become a critical environmental challenge, motivating the need for advanced technologies capable of accurately identifying plastic materials for efficient waste management and large-scale monitoring. Hyperspectral imaging (HSI), particularly in the short-wave infrared (SWIR) range, provides rich spectral–spatial information that enables precise discrimination between visually similar materials based on their chemical composition. However, the high dimensionality and complexity of hyperspectral data present major challenges for automated analysis. This thesis investigates two separate deep learning–based systems for plastic detection using SWIR hyperspectral imaging. The first system is fully unsupervised and relies on a 3D autoencoder to learn compact spectral–spatial representations, followed by the Deep Embedded Clustering (DEC) algorithm to group hyperspectral patches into meaningful clusters without requiring annotated data. This system aims to explore the intrinsic structure of the data and assess its ability to reveal distinct plastic categories. The second system is purely supervised and built around a 3D-CNN architecture designed to perform pixel-level plastic classification using manually defined labels. This model focuses on learning discriminative spectral–spatial features directly from labeled data to achieve accurate detection performance. Both systems were evaluated independently on SWIR hyperspectral dataset. The unsupervised system demonstrated coherent clustering behavior and meaningful latent-space organization, while the supervised classifier achieved high accuracy, strong generalization, and stable convergence. The results highlight the complementary value of studying both approaches separately and provide a deeper understanding of how hyperspectral learning techniques can be exploited for plastic detection in real-world scenarios. |
| Sommaire : |
General Introduction 7 1 Stateoftheart 9 1.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2 Hyperspectral Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2.2 TheHyperspectraldatacube . . . . . . . . . . . . . . . . . . . . 11 1.2.3 DataAcquisitionStrategies . . . . . . . . . . . . . . . . . . . . . 12 1.3 SpectralSignaturesofMaterials . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.1 SpectralReflectance . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.2 SpectralDistinctionofPlastics . . . . . . . . . . . . . . . . . . . 15 1.3.3 RoleofNIRandSWIRinPlasticDetection . . . . . . . . . . . . 15 1.4 MainTasksinHSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.4.1 InformationExtractionTasks . . . . . . . . . . . . . . . . . . . . 17 1.4.2 DataTransformationTasks . . . . . . . . . . . . . . . . . . . . . 18 1.4.3 DataEnhancementandCorrectionTasks . . . . . . . . . . . . . 19 1.4.4 TemporalAnalysisTasks. . . . . . . . . . . . . . . . . . . . . . . 20 1.5 BenefitsandChallengesofUsingHSI . . . . . . . . . . . . . . . . . . . . 20 1.5.1 Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 1.5.2 Challenges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.6 SmartHSIsystems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.6.1 Artificial Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.6.2 MachineLearning . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.6.3 DeepLearning(DL) . . . . . . . . . . . . . . . . . . . . . . . . . 26 1.6.4 AI inHyperspectral Imaging . . . . . . . . . . . . . . . . . . . . 26 1.7 RelatedWorksonPlasticDetectionUsingHSI . . . . . . . . . . . . . . . 27 1.8 ApplicationsofHSI inPlastics . . . . . . . . . . . . . . . . . . . . . . . 29 1.9 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2 AnalysisandSpecifications 31 2.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.2 FunctionalRequirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.3 Non-FunctionalRequirements . . . . . . . . . . . . . . . . . . . . . . . . 32 2.4 DataRequirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 2.4.1 HyperspectralDataSource . . . . . . . . . . . . . . . . . . . . . 34 2.4.2 LabelsorAnnotationsforPlasticTypes . . . . . . . . . . . . . . 35 2.4.3 PreprocessingRequirements . . . . . . . . . . . . . . . . . . . . . 35 2.4.4 AugmentationandDataBalancing . . . . . . . . . . . . . . . . . 36 2 2.4.5 DataValidationandQualityAssurance . . . . . . . . . . . . . . 36 2.5 SystemSpecifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 2.5.1 TechnicalRequirements . . . . . . . . . . . . . . . . . . . . . . . 36 2.5.2 FunctionalBlocks. . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2.6 ToolsandTechnologyChoices . . . . . . . . . . . . . . . . . . . . . . . . 38 2.6.1 ProgrammingLanguageandLibraries . . . . . . . . . . . . . . . 38 2.6.2 HardwareandPlatforms . . . . . . . . . . . . . . . . . . . . . . . 40 2.6.3 DatasetSources. . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.6.4 EvaluationMetrics . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.7 RiskAnalysisandLimitations . . . . . . . . . . . . . . . . . . . . . . . . 44 2.8 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3 Contributions:DesignandImplementation 46 3.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 3.2 UnsupervisedLearning-basedModel . . . . . . . . . . . . . . . . . . . . 47 3.2.1 ArchitectureOverview . . . . . . . . . . . . . . . . . . . . . . . . 47 3.2.2 DetailedDesign. . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.3 SupervisedLearning-basedModel . . . . . . . . . . . . . . . . . . . . . . 54 3.3.1 ArchitectureOverview . . . . . . . . . . . . . . . . . . . . . . . . 54 3.3.2 DetailedDesign. . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 3.4 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 3.4.1 HyperspectralDataAcquisition . . . . . . . . . . . . . . . . . . . 62 3.4.2 HyperspectralDataStructureandPreprocessing . . . . . . . . . 63 3.4.3 PlasticMaterialsIncluded . . . . . . . . . . . . . . . . . . . . . . 64 3.5 UMLDiagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.5.1 classdiagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 3.5.2 sequencediagram. . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.6 DevelopmentToolsandLibraries . . . . . . . . . . . . . . . . . . . . . . 71 3.7 TrainingConfiguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 3.8 ResultsAnalysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.8.1 UnsupervisedLearning-basedModel . . . . . . . . . . . . . . . . 75 3.8.2 SupervisedLearning-basedModel . . . . . . . . . . . . . . . . . . 82 3.9 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 3.10Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 GeneralConclusion 93 |
| Type de document : | Mémoire master |
Disponibilité (1)
| Cote | Support | Localisation | Statut |
|---|---|---|---|
| MINF/967 | Mémoire master | bibliothèque sciences exactes | Consultable |
