Preeyanuch Srichola

The success of large-scale deep learning models in remote sensing tasks has been transformative, enabling significant advances in image classification, object detection, and image–text retrieval. However, their computational and memory demands pose challenges for deployment in resource-constrained environments. Knowledge distillation (KD) alleviates these issues by transferring knowledge from a strong teacher to a student model, which can be compact for efficient deployment or architecturally matched to improve accuracy under the same inference budget. In this paper, we introduce Hierarchical Multi-Segment Knowledge Distillation (HIMS_KD), a multi-stage framework that sequentially distills knowledge from a teacher into multiple assistant models specialized in low-, mid-, and high-level representations, and then aggregates their knowledge into the final student. We integrate feature-level alignment, auxiliary similarity-logit alignment, and supervised loss during distillation. Experiments on benchmark remote sensing datasets (RSITMD and RSICD) show that HIMS_KD improves retrieval performance and enhances zero-shot classification; and when a compact student is used, it reduces deployment cost while retaining strong accuracy.

Knowledge distillation (KD) enables the training of lightweight yet effective models, particularly in the visual domain. Meanwhile, reinforcement learning (RL) facilitates adaptive learning through environment-driven interactions, addressing the limitations of KD in handling dynamic and complex tasks. We propose a novel two-stage framework integrating Knowledge Distillation with Reinforcement Learning (KDRL) to enhance model adaptability to complex data distributions, such as remote sensing and medical imaging. In the first stage, supervised fine-tuning guides the student model using logit and feature-based distillation. The second stage refines the model via RL, leveraging confidence-based and cluster alignment rewards while dynamically reducing reliance on task loss. By combining the strengths of supervised knowledge distillation and reinforcement learning, KDRL provides a comprehensive approach to address the dual challenges of model efficiency and domain heterogeneity. A key innovation is the introduction of auxiliary layers within the student encoder to evaluate and reward the alignment of the characteristics with the teacher’s cluster centers, promoting robust feature learning. Our framework demonstrates superior performance and computational efficiency across diverse tasks, establishing a scalable design for efficient model training. Across remote sensing benchmarks, KDRL boosts the lightweight CLIP/ViT-B-32 student to 69.51% zero-shot accuracy on AID and 80.08% on RESISC45; achieves state-of-the-art cross-modal retrieval on RSITMD with 67.44% (I→T) and 74.76% (T→I) at R@10; and improves DIOR-RSVG visual-grounding precision to 64.21% at Pr@0.9. These gains matter in real deployments by reducing missed targets and speeding analyst search on resource-constrained platforms.

This study presents a custom-built, portable multispectral imaging (MSI) system integrated with computer vision for sodium nitrite detection via the Griess reaction on paper-based substrates. The MSI system was used to investigate the absorption characteristics of sodium nitrite at concentrations from 0 to 10 ppm across nine spectral bands spanning 360–940 nm on para-aminobenzoic acid (PABA) and sulfanilamide (SA) substrates. Upon forming azo dyes with N-(1-naphthyl) ethylenediamine (NED), the PABA and SA substrates exhibited strong absorption near 545 nm and 540 nm, respectively, as measured by a spectrometer. This agrees with the 550 nm MSI images, in which higher sodium nitrite concentration regions appeared darker due to increased absorption. A concentration-correlation analysis was conducted for each spectral band. The normalized difference index (NDI), constructed from the most and least correlated bands at 550 nm and 940 nm, showed a stronger correlation with sodium nitrite concentration than the single best-performing band for both substrates. The NDI increased the coefficient of determination (R²) by approximately 19.32% for PABA–NED and 19.89% for SA–NED. This improvement was further confirmed under varying illumination conditions and through comparison with a conventional smartphone RGB imaging approach, in which the MSI-based NDI showed substantially superior performance. The enhancement is attributed to improved contrast, illumination normalization by the NDI, and the narrower spectral bands of the MSI compared with RGB imaging. In addition, the NDI framework enabled effective image segmentation, classification, and visualization, improving both interpretability and usability and providing a practical guideline for developing more robust models with larger training datasets. The proposed MSI system offers strong advantages in portability, sub-minute acquisition time, and operational simplicity, enabling rapid, on-site, and non-destructive chemical analysis.

The aim of this study is to investigate the effect of hydrogen bonds in natural cellulose on luminescent properties, specifically focusing on controlling optical characteristics by tuning hydrogen bond interactions through the incorporation of light scattering from gold nanoparticles (AuNPs). The results from gold nanoparticles are that all samples exhibited strong plasmonic scattering within the 350–800 nm range, albeit with varying intensities. This study demonstrates that cellulose combined with AuNPs is highly effective for light scattering applications due to the reflective properties of cellulose and the surface plasmon resonance (SPR) effects of AuNPs. The enhancement of the autofluorescence signal increased about 2X relative to the Agave-AuNP sample and 67% higher than pure agave cellulose autofluorescence. Moreover, a quenching effect was observed in the mixture of cellulose, C₉H₂₃NO₃Si, (APTES) and AuNPs attributed to hydrogen bond interactions, which diminished the light scattering properties. To understand the autofluorescence properties of cellulose and its interaction with metal nanoparticles, these composite materials are promising candidates for novel applications in bioimaging, sensing, and optoelectronic devices.

Cellulose-based paper products derived from agro-industrial waste have attracted considerable interest due to their potential in sustainable material development. In this study, bacterial cellulose (BC) residue from the food and beverage industry was employed as a reinforcing agent to fabricate high-performance paper composites by blending with eucalyptus pulp (EP) at various ratios and basis weights. These papers were coated with a cationic modified starch solution (MS) using a rod coater, followed by hot pressing. Mechanical strengths (TAPPI Standard), water resistance (Cobb test and water contact angle), and air permeability (ASTM D737) were evaluated to assess material performance. The results showed that incorporating 50 wt% BC produced paper with outstanding mechanical performance, characterized by a high tensile index and excellent tear resistance. The application of the MS coating significantly boosted water resistance and air barrier performance, underscoring the effectiveness of this approach in creating high-performance paper materials. The resulting coated composites demonstrated excellent mechanical strength and barrier properties, positioning them as promising candidates for filtration applications such as personal protective face mask membranes.

This study investigates the effect of bases NaOH and KOH on okara, the soybean residue, in conventional pulping, based on 136 pulping conditions used as a dataset for random forest regression and gradient boosting predictive models. Okara CMC was formed and identified using Fourier-transform infrared spectroscopy (FTIR) to demonstrate a wide range of applications comparable to commercial CMC, with a low degree of substitution. The quality of okara pulp after basic pulping was analyzed based on the extracted cellulose yield and remaining protein content. The optimized pulping condition was a mixture of NaOH and KOH at a 30% concentration, resulting in an extracted cellulose yield of 24.5 wt% and a remaining protein content of 25.1%. The obtained okara pulp was converted into okara CMC with a controllable degree of substitution. The implemented dataset was used to train two predictive models: random forest regression and gradient boosting, to forecast key parameters for pulping (NaOH, KOH, AQ, and H₂O). Both models demonstrated excellent prediction performance, with R² values of 0.94 and 0.89, respectively, and showed similar residuals and predicted values. The close clustering of residuals around zero, along with the sharp and narrow curves observed, indicates that both the random forest and gradient boosting models provide precise and reliable predictions. The localized deviations observed in the residuals suggest that these models effectively capture detailed patterns in the data, leading to minimized prediction errors within specific ranges.

Since the late 2010-s and early 2020s, people around the world have not only encountered the pandemic crisis, but also in some places, they have had to deal with serious levels of air pollution. Personal protective equipment is essential to protect from microorganisms or fine particulate matter. Consequently, this study aimed to develop a silk face covering that could meet the international requirements with the addition of having an antibacterial property. The developed silk face covering consisted of three layers; the outer layer was water-repellent mulberry silk, the inner layer was oil-repellent eri silk, while the middle layer was cotton fabric coated with biosynthesized silver nanoparticles from mangosteen peels mixed with Andrographis paniculata extract. A biodegradable bacterial cellulose filter made of nata de coco waste was also prepared to improve filtration efficiency. It was found that the silver nanoparticles extracted from mangosteen peel and Andrographis paniculata inhibited S. aureus and E. coli by more than 99.9%, even after 20 washing cycles. The performance properties of the silk barrier face covering met Level I of the ASTM F3502-2021 standard, as well as being comfortable for public use.

Textile waste has emerged as a critical global challenge, with improper disposal practices leading to adverse environmental consequences. In response to this pressing issue, there is growing interest in recycling textile waste containing cellulose as an alternative approach to reducing the impact of industrial waste on the environment. The objective of this research is to investigate the extraction and characterization of nanocellulose from polyester–cotton textile waste as a potential solution to address the growing concerns of waste management in the textile industry. To obtain nanocellulose, a comprehensive process involving alkaline sodium hydroxide (NaOH) treatment of the polyester–cotton textile (35% PET and 65% cotton) was employed, resulting in average yield percentages ranging from 62.14% to 71.21%. To achieve the complete hydrolysis of PET polyester in the blends, second hydrolysis was employed, and the optimized condition yield cotton fiber was 65.06 wt%, relatively close to the theoretical yield. Subsequently, the obtained cellulosic material underwent an acid hydrolysis process using 70 percent (v/v) sulfuric acid (H₂SO₄) solution at 45 °C for 90 min, resulting in nanocellulose. Centrifugation at 15,000 rpm for 15 min facilitated the separation of nanocellulose from the acid solution and yielded 56.26 wt% at optimized conditions. The characterization of the nanocellulose was carried out utilizing a comprehensive array of techniques, including absorption, transmission, and reflection spectra, and Fourier transform infrared. The characterization results provide valuable insights into the unique properties of nanocellulose extracted from textile waste. In this research, the obtained nanocellulose was mixed with PVA and silver nanoparticle to form biodegradable film composites as the reinforcement. In comparison, biodegradable film of PVA:nanocellulose 9.5:0.5 with silver nanoparticle 0.3 wt% and glycerol as a plasticizer exhibits better tensile strength (2.37 MPa) and elongation (214.26%) than the PVA film with normal cellulose. The prepared biodegradable film was homogeneous and had a smooth surface without the internal defect confirmed by the CT scan. This result opens avenues for enhancing the quantities of eco-friendly film composites, potentially replacing conventional plastic films in the future.

Textile waste usually ends up in landfills and causes environmental pollution. In this study, pretreatment methods for textile recycling, including autoclaving, freezing alkali/urea soaking, and alkaline pretreatment, were applied to textile waste with various cotton/polyester blending ratios. The best condition for enzymatic hydrolysis was a 60/40 textile waste blend of cotton/polyethylene terephthalate (PET) with a reusable chemical pretreatment (15% NaOH) at 121 °C for 15 min. The hydrolysis of pretreated textile waste by cellulase was optimized using response surface methodology (RSM) based on central composite design (CCD). The optimized conditions were 30 FPU/g of enzyme loading and 7% of substrate loading, which resulted in a maximum observed value of hydrolysis yield at 89.7%, corresponding to the predicted value of 87.8% after 96 h of incubation. The findings of this study suggest an optimistic solution for textile waste recycling.