Non-destructive quantification of sea lettuce in laver using hyperspectral imaging with hybrid spectral feature selection techniques
Jong‐Jin Park, Seulki Park, Dae-Yong Yun, Gyuseok Lee, Sang Seop Kim, Kee‐Jai Park, Jeong‐Ho Lim, Jeong‐Hee Choi, Jeong‐Seok Cho
Abstract
The quality of laver is significantly affected by adulteration with sea lettuce ( Ulva lactuca ), a green seaweed that adheres to Pyropia nets during cultivation and adversely impacts productivity and quality. Acid treatment agents are commonly utilized; however, residual sea lettuce may persist in the final product if treatment is insufficient. Traditional detection methods, such as sensory evaluation, are susceptible to human error, time-consuming, and inefficient, while DNA sequencing is ineffective for processed laver due to DNA degradation. Given these limitations, non-destructive technologies are garnering interest in seafood quality assessment. This study evaluates the potential of hyperspectral imaging for detecting sea lettuce in laver. Hypercubes collected in two spectral ranges (visible/near-infrared (VIS/NIR) and short-wave infrared (SWIR)) were utilized to establish a partial least squares regression (PLSR) model for quantification. Characteristic wavelengths were selected using competitive adaptive reweighted sampling (CARS), uninformative variable elimination (UVE), and their hybrid methods (CARS-UVE, UVE-CARS). Model efficiency and robustness improved with spectral feature selection. For raw laver, UVE-CARS achieved the highest R p 2 (0.86) with 14.3% of full wavelengths in VIS/NIR, while for dried laver, SWIR with CARS-UVE yielded R p 2 (0.90) using 18.3% of full wavelengths. This study addresses a critical gap in seafood quality control by demonstrating that hyperspectral imaging enables non-destructive, efficient quantification of sea lettuce contamination in laver, contributing to improved industry standards. • Hyperspectral imaging enables the detection of sea lettuce in raw and dried laver. • Appropriate spectral ranges and feature selection varied by laver processing state. • VIS/NIR combined with UVE-CARS effectively quantified sea lettuce in raw laver • SWIR with CARS-UVE is suitable for sea lettuce quantification in dried laver