We present a real-valued intensity transmission matrix (RVITM) framework that enables phase-free structured illumination control through multimode fibers, overcoming long-standing challenges of modal scrambling from dispersion and bending. Unlike previous TM-based approaches that require complex-field measurements or interferometric detection, our method relies solely on 16 × 16 Hadamard intensity inputs, enabling deterministic synthesis of circularly symmetric illumination patterns that remain stable under fiber bending. To the best of our knowledge, this is the first demonstration of RVITM-based ring illumination through a multimode fiber for Raman spectroscopy and the first proof-of-concept application to inverse spatially offset Raman spectroscopy (iSORS). Using a 3 mm-radius ring pattern for iSORS of a two-layer phantom, the subsurface-to-surface Raman peak ratio was enhanced by 1.5 compared with point illumination, confirming improved depth sensitivity without added system complexity. RVITM offers a calibration-friendly and mechanically robust route to fiber-delivered structured illumination, advancing minimally invasive spectroscopy and imaging applications.

Many IoT systems require deadline-constrained workflow scheduling, where missed deadlines can have serious consequences. Scheduling such IoT workflows in Fog-Cloud environments is challenging due to resource heterogeneity and the variability in workflow patterns and deadlines. Existing approaches, including heuristic and meta-heuristic algorithms, often fail to reliably satisfy deadline constraints while simultaneously minimizing the cost associated with the computational resources used for executing workflows. This paper introduces the Internet of Things Genetic Algorithm with Selective Repair under Combined Criteria (IoTGA-SRC2) to effectively tackle these challenges. IoTGA-SRC2 introduces a novel selection mechanism that prioritizes solutions based on deadline violations and execution costs. It also features an innovative repair method, which can systematically detect infeasible solutions, perform a root cause analysis to identify the key factors causing deadline violations, and reallocate critical tasks using a multi-criteria method. By properly managing delays caused by execution time, communication time, and waiting time, IoTGA-SRC2 can consistently satisfy deadline constraints across a wide range of problem configurations. Extensive experiments demonstrate that IoTGA-SRC2 consistently outperforms multiple state-of-the-art methods in reducing execution costs while adhering to stringent deadline constraints, making it a valuable choice for various real-world applications in heterogeneous IoT-Fog-Cloud computing environments.

Bone-related disorders, including osteoporosis, lead to decreased bone density and altered microarchitecture, significantly increasing the risk of fragility fracture. There is an urgent need for innovative early detection methods, as conventional techniques like dual-energy x-ray absorptiometry offer limited insights into bone quality, often overlooking critical microstructural changes. In this study, we present a novel dual-wavelength inverse spatially offset Raman spectroscopy (DWiSORS) system specifically designed to enhance the assessment of transcutaneous bone signals. This innovative approach utilises ring illumination to deliver adequate power while minimising the fluorescence background, and dual-wavelength excitation to capture a broader molecular profile (spectral range). We introduce a novel metric, enhancement-to-noise ratio (ENR), which provides a quantitative and robust strategy to identify the optimal offset for accurate assessment at specified depths, addressing a significant limitation in existing in-vivo SORS methodologies. The metric efficacy was validated through in-vivo measurements on ten healthy volunteers, where the ENR was analysed against increasing source-detector offsets, with tissue thickness determined by ultrasound imaging. Notably, optimal signal-to-noise ratio for bone signals were achieved at offsets of 7 or 9 mm, even beneath 10-14 mm of overlaying tissue, reinforcing increasing the offset beyond this threshold does not necessarily enhance bone signals detection. In-vivo measurements demonstrate the technical feasibility and robustness of the proposed DWiSORS framework for depth-resolved bone Raman measurements. This study indicates the potential of Raman spectroscopy to complement conventional imaging techniques and support future approaches for non-invasive assessment and longitudinal monitoring of bone-related disorders.

Polarimetric imaging offers a label-free advantage for pathology because it captures scattering-related tissue information without chemical staining, and recent studies suggest that unstained polarization signals can support tissue recognition. Yet prior Stokes-to-bright-field studies remain largely limited to stained inputs and GAN-based translation. This study presents a diffusion-prior framework that generates H E-like bright-field images directly from unstained Stokes images acquired by a single-shot division-of-focal-plane system. Compared with stained-slice translation, this setting is substantially more difficult because the polarimetric signal is weaker and the post-staining references are only physically corresponding rather than strictly aligned. To address this problem, the framework combines stain normalization with a compact single-step generator constrained by adversarial, cycle-consistency, perceptual, and identity losses. Experiments show stable optimization, consistent quantitative and qualitative improvements over multiple representative baselines, including CycleGAN and more recent generative methods. These results support diffusion-prior Stokes translation as a practical direction for clinically readable virtual H E reconstruction from unstained tissue.

Autoscaling is an important technique for cloud computing that dynamically adjusts resources allocated to cloud applications in response to fluctuating user requests to maintain Quality of Service (QoS) and adhere to a given budget. Recent advancements in Deep Reinforcement Learning (DRL) have shown promise in achieving effective autoscaling approaches. However, prior DRL-based approaches struggle to simultaneously consider the spatial dependencies within an application and the changing historical workload patterns, limiting their ability to make accurate scaling decisions. Moreover, existing approaches lack the fine-grained resource adjustment, leading to suboptimal autoscaling performance. To address these limitations, we propose a new DRL-based autoscaling approach with a novel spatial-temporal autoscaling policy, which jointly captures spatial and temporal features of cloud applications by Graph Neural Networks and Transformers. Meanwhile, this policy enables fine-grained resource adjustment. Extensive experiments on real-world user request traces show that the proposed approach significantly outperforms existing state-of-the-art methods, achieving up to a 78.23% reduction in mean response time without violating the cost budget.

Objective The accurate and efficient identification of marine microalgae in historical natural water samples is of paramount importance for reconstructing past phytoplankton dynamics, assessing long-term ecological trends, and improving early-warning systems for harmful algal blooms (HABs). However, the vast archives of preserved water samples-typically fixed with formaldehyde for morphological stabilization-remain underutilized due to significant challenges in automated analysis. Formaldehyde fixation induces structural and optical alterations in microalgal cells, including changes in refractive index, cell wall rigidity, and light-scattering properties, which severely degrade the performance of conventional image-based identification methods. While manual microscopy remains the gold standard, it is prohibitively time-consuming, subjective, and impractical for large-scale retrospective studies. Emerging techniques such as flow cytometry, hyperspectral imaging, and fluorescence-based classification either lack species-level resolution or are incompatible with degraded, fixed specimens stored over extended periods. Therefore, there is a critical and unmet need for a robust, label-free, and automation-ready methodology capable of extracting discriminative features from formaldehyde-preserved microalgae. To address this gap, this study pioneers an integrative approach that synergistically combines Mueller matrix polarimetry-a modality highly sensitive to subcellular structural anisotropy and morphology-with deep learning architectures, enabling high-fidelity, automatic recognition of microalgal species in historically archived samples. Methods A custom-built transmission-mode Mueller matrix microscope operating at a wavelength of 532 nm was employed to capture complete 4 & times;4 polarization response images of seven representative cultured microalgal species (Skeletonema costatum, Pseudonitzschia pun gens, Chlorella vulgaris, Dicrateria zhanjiangensis, Isochrysis galbana, Phaeodactylum tricornutum, and Thalassiosira weissflogii), under both unfixed (live) and formaldehyde-fixed conditions. This yielded two distinct, rigorously annotated training datasets: Dataset A (fixed) and Dataset B (unfixed), each comprising hundreds of single-cell instances (Table 1). A modified Faster R-CNN object detection framework was developed, featuring a 16-channel input layer to accommodate the full Mueller matrix (flattened into 16 polarization channels) and a ResNet-50 backbone fine-tuned for microscopic biological targets. Two parallel classifiers-Classifier A (trained on Dataset A) and Classifier B (trained on Dataset B)-were trained and validated. These models were then deployed to identify Skeletonema costatum and Pseudo-nitzschia pun gens in two sets of real-world historical natural water samples collected in January and May 2025, respectively, and preserved in formaldehyde for approximately 4 months and 15 days prior to analysis. Ground-truth annotations were meticulously established across 463 fields of view by expert taxonomists. To further investigate the temporal evolution of fixation-induced polarization changes, time-resolved Mueller matrix measurements were conducted on three model species (Chlorella, Dicrateria zhanjiangensis, and Isochrysis galbana) immediately after fixation (at minute-scale intervals for 120 minutes) and daily over a 10-day period. Principal component analysis (PCA) was applied to the 16-dimensional Mueller vector data to quantify and visualize the dominant modes of polarization variation over time (Fig. 7, Table 3). Results and Discussions The proposed Mueller matrix -deep learning pipeline achieved good performance in controlled settings: Classifier A attained an average per-species accuracy of 93.8% on the seven cultured species, surpassing Classifier B (92.3% ) and dramatically outperforming a baseline intensity-only classifier (76.4%) (Fig. 4, Fig. 8). When applied to historical natural samples, Classifier A demonstrated superior generalization capability: it correctly identified target diatoms with average accuracies of 81.0% in the January samples (4-month fixation) and 90.5% in the May samples (15-day fixation), whereas Classifier B yielded lower accuracies of 69.2% and 87.3 %, respectively (Fig. 5, Table 2). This consistent advantage underscores the critical importance of sample alignment-training on fixed samples better matches the biophysical state of archived specimens. Notably, the system successfully resolved chain-forming diatoms into individual cells during detection, achieving a remarkable 88.8% accuracy on the May dataset (Fig. 6). The PCA results revealed that formaldehyde fixation triggers rapid and profound changes in microalgal polarization signatures: the first principal component analysis (PCA1), accounting for >70% of total variance, exhibited sharp shifts within the first 10-30 min post-fixation, followed by gradual stabilization by 24 h (Fig. 7). Over the subsequent 10 d, PCA1 values remained remarkably stable, indicating that the polarization phenotype reaches a quasi-equilibrium state (Table 3). The significantly higher variance observed in the minute-scale phase versus the day-scale phase confirms that the most dramatic biophysical transformations occur immediately after chemical fixation. These findings provide a mechanistic explanation for the superior performance of Classifier A and establish a practical protocol: Mueller matrix data for training should be acquired no earlier than 24 h post-fixation to ensure representativeness and stability. Conclusions This work establishes Mueller matrix polarimetry coupled with deep learning as a powerful, reliable, and scalable solution for the automated identification of microalgae in formaldehyde-fixed historical water samples-a longstanding challenge in marine ecology and environmental monitoring. The method achieves high accuracy (up to 90.5% ) even in historical natural matrices and significantly outperforms conventional intensity-based approaches by leveraging rich polarization-encoded structural information that persists despite fixation-induced degradation. The study also demonstrates that while formaldehyde alters microalgal optical properties, these changes stabilize within one day, enabling the creation of consistent and transferable training datasets. The findings not only validate the feasibility of unlocking decades of archived samples for retrospective phytoplankton analysis but also pave the way for integrating polarization-aware AI into next-generation HAB monitoring platforms. Future work will expand the species library, explore multi-wavelength Mueller imaging, and deploy the system in field-deployable configurations for near-real-time ecological assessment.

Microservice architecture has become a dominant paradigm in application development due to its advantages of being lightweight, flexible, and resilient. Deploying microservice applications in the container-based cloud enables fine-grained elastic resource allocation. Autoscaling is an effective approach to dynamically adjust the resource provisioned to containers. However, the intricate microservice dependencies and the deployment scheme of the container-based cloud bring extra challenges of resource scaling. This article proposes a novel autoscaling approach named HGraphScale. In particular, HGraphScale captures microservice dependencies and the deployment scheme by a newly designed hierarchical graph neural network, and makes effective scaling actions for rapidly changing user requests workloads. Extensive experiments based on real-world traces of user requests are conducted to evaluate the effectiveness of HGraphScale. The experiment results show that the HGraphScale outperforms existing state-of-the-art autoscaling approaches by reducing at most 80.16% of the average response time under a certain VM rental budget of application providers.

The increasing complexity and variety of IoT systems require the integration of multiple services to meet a wide range of user needs. This paper addresses the challenge of multi-objective IoT service composition with replication problem by considering multiple Quality of Service (QoS) metrics such as response time and the number of selected service instances. We propose a new Memetic NSGA-II algorithm with Bottleneck-driven Local Search (MNSGA2-BLS) to effectively solve this difficult problem. By integrating genetic operations, clustering-based refinement, and a bottleneck-driven Estimation of Distribution Algorithm for local search, MNSGA2-BLS identifies and optimizes critical service instances causing QoS bottlenecks. This method leverages Pareto-optimal solutions to guide the local search refinement process, enhancing convergence and solution quality. Experimental results across various benchmark cases demonstrate that MNSGA2-BLS can outperform NSGA-II and several state-of-the-art algorithms, achieving superior results in both the hyper-volume and inverse generational distance metrics. This highlights the potential of MNSGA2-BLS to provide efficient and effective composite IoT services while addressing trade-offs between competing QoS objectives.

Microalgae are pivotal to aquatic ecosystems, yet differentiation is constrained by limited cell data. A method termed multiple polarization imaging of microalgae at different postures (DP-MPI) is proposed and demonstrated by species classification and individual recognition. A herringbone microchannel is designed to tumble the microalgal cells and then change their postures. A conceptual setup of multiple polarization imaging is built to obtain a series of polarization images of the cells at different postures with high throughput while they flow through the microchannel. 12 representative species comprising 8,716 individual cells are measured, yielding 178,556 polarization images (up to 50 per cell). Three schemes, including the usage of the single polarization image and the multiple polarization images, are compared to indicate DP-MPI's performance. With deep learning, a macro-average species classification accuracy of 98.8 % is achieved by DP-MPI, representing a 33.4 % improvement over the state-of-the-art method, and arbitrary individual recognition up to 96.5 % is attained, which is almost impossible for existing methods. Influencing factors, including image number and illuminating polarization states, are analyzed to guide the application. Compared with multiple intensity images, DP-MPI's average accuracy is 20.8 % higher. These findings demonstrate the method's power as a promising tool for aquatic environmental monitoring.

We propose a physics-based unsupervised segmentation method. By quantifying the wavelength-induced centroid shift of 44-dimensional polarization features, the method segments key microstructures, including cell nuclei, collagen fibers, and cytoplasm.

Microplastics (MPs), as global emerging contaminants, pose a persistent threat to ecosystems and human health. However, current MP differentiation techniques are typically time-consuming and labor-intensive, limiting their applicability for environmental monitoring. This paper proposes a high-throughput MP differentiation method called pixel-based polarization classification (PBPC). The setup can acquire backscattered Mueller matrix images of multiple MPs. For each pixel of a single MP, 59 polarization parameters are derived from its Mueller matrix to represent a pixel polarization vector (PPV). A total of 20 types of MPs are measured in the data set, with at least 1 million PPVs for each type. Three different machine learning classifiers are trained respectively, and the optimal one achieves an accuracy of 90.24% in PPV classification. The results are visualized as the region classification image, and the pixel classification proportions of each MP are further evaluated. In this work, the high-throughput capability of PBPC to differentiate MPs with diverse morphologies is demonstrated by standard samples. For environmental MP samples, the detection results remain consistent with mu-FTIR, validating the robustness and generalization of PBPC. Moreover, the characterization of PPVs is analyzed, and the impact of abnormal pixels caused by imaging overexposure is quantitatively assessed. A detailed differentiation of two MPs with varying densities, HDPE and LDPE, highlights PBPC's sensitivity to subtle structural differences. This work demonstrates PBPC's potential as a promising tool for high-throughput MP differentiation, which would facilitate environmental monitoring and MP pollution assessment.

Mueller matrix imaging encodes rich pathological microstructural information within a 16-dimensional polarization space. Polarization super-pixels (PSPs), obtained by grouping Mueller matrix pixels with similar polarization characteristics, provide an effective representation for reducing data dimensionality while preserving essential polarization features, enabling the characterization of key microstructural components such as fibrous structures and cancer cell nuclei. The development of digital pathology relies on large-scale, high-quality annotated datasets for model training. However, manual annotation of whole slide images (WSIs) is labor-intensive and time-consuming, as pathologists must repeatedly identify and label similar pathological structures. To address this limitation, we propose a polarization feature template (PFT)-based framework for accurate and efficient annotation. In this approach, a limited number of well-annotated samples are used to assign weights to PSPs and construct representative PFTs, which are subsequently applied to annotate unknown samples based on their polarization-encoded pathological characteristics. Experimental results demonstrate that the PFT constructed from pixel-level annotations achieves complete separation between squamous cell carcinoma and adenocarcinoma, while the PFT derived from ROI-level annotations attains an AUROC of 0.94 for subtype discrimination. Overall, the proposed framework shows the potential to reduce annotation burden, improve annotation accuracy and efficiency, and facilitate the construction of high-quality annotated datasets for digital pathology applications.

Workflow scheduling in cloud computing demands the intelligent allocation of dynamically arriving, graph-structured workflows with varying deadlines onto ever-changing virtual machine resources. However, existing deep reinforcement learning (DRL) schedulers remain limited by rigid, single-path inference architectures that struggle to handle diverse scheduling scenarios. We introduce $\textbf{DEFT}$ ($\textbf{D}$eadline-p$\textbf{E}$rceptive Mixture-o$\textbf{F}$-Exper$\textbf{t}$s), an innovative DRL policy architecture that leverages a specialized mixture of experts, each trained to manage different levels of deadline tightness. To our knowledge, DEFT is the first to introduce and validate a Mixture-of-Experts architecture for dynamic cloud workflow scheduling. By adaptively routing decisions through the most appropriate experts, DEFT is capable of meeting a broad spectrum of deadline requirements that no single expert can achieve. Central to DEFT is a $\textbf{graph-adaptive}$ gating mechanism that encodes workflow DAGs, task states, and VM conditions, using cross-attention to guide expert activation in a fine-grained, deadline-sensitive manner. Experiments on dynamic cloud workflow benchmarks demonstrate that DEFT significantly reduces execution cost and deadline violations, outperforming multiple state-of-the-art DRL baselines.