Integration of Digital Colorimetry and Machine Learning for the Evaluation of CdZnSeS Quantum Dots

Tan E., Snee P. T., Akgündoğdu A.

20th Nanoscience and Nanotechnology Conference, NanoTR-20, İzmir, Türkiye, 26 - 28 Ağustos 2026, ss.265, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: İzmir
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.265
  • İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet

Özet

Quantum dots (QDs) are semiconductor nanocrystals with sizes typically ranging from 2 to 10 nm [1]. Owing to their pronounced sizedependent optical and chemical properties, their tunable color variations have attracted significant attention in colorimetric sensing

applications. However, the quantitative interpretation of colorimetric responses remains challenging and often requires complementary

spectroscopic analyses. In this study, CdZnSeS quantum dots were synthesized via the hot-injection method under systematically varied

precursor ratios and reaction temperatures. Their structural and optical properties were investigated using X-ray diffraction (XRD), UV–Vis

spectroscopy, and photoluminescence (PL) spectroscopy. The results showed that the quantum yield (%) of the QDs varied up to 85%,

depending on their composition. Digital colorimetric data were extracted from UV-illuminated images of the synthesized QDs, and Euclidean

color difference (ΔE) values were calculated to quantify perceptual color variations via RGB analysis. To evaluate the predictive potential of

image-derived color features, a hybrid Adaptive Neuro-Fuzzy Inference System (ANFIS) was developed to estimate PL intensity using RGB

values and ΔE as input parameters [2,3]. To improve model robustness and generalization performance, Gaussian noise-based data

augmentation was applied to the original dataset consisting of 30 samples, expanding it to a total of 150 samples.

Correlation analysis revealed that the green channel exhibited the highest positive correlation with PL intensity (r = 0.51). Furthermore, ΔE

displayed strong correlations with both the red and green channels, indicating its effectiveness in representing combined color variations and

overall optical behavior. The proposed ANFIS model demonstrated strong predictive capability for estimating PL intensity from image-derived

color features. The model achieved a Pearson correlation coefficient of R = 0.917 and a coefficient of determination (R²) of 0.749 on unseen

test samples. The incorporation of output constraints enabled the generation of physically meaningful predictions while effectively capturing

dynamic variations in PL intensity. These findings demonstrate that image-derived colorimetric descriptors can serve as rapid and low-cost

proxies for optical measurements, highlighting the potential of integrating digital colorimetry, machine learning, and fuzzy intelligence into

next-generation quantum dot sensing platforms.