Enhancing fruit SSC detection accuracy via a light attenuation theory-based correction method to mitigate measurement orientation variability

Abstract

Nondestructive online detection and sorting for fruit quality has gradually attracted attention in the global agro-product industry. However, the detection accuracy is influenced by many factors, such as fruit orientation, fruit shape, and environmental fluctuations. This study aimed to explore the impact of measurement orientation variation on spectra and soluble solids content (SSC) detection in apples and propose a correction method to mitigate the effect. Firstly, the visible/near-infrared (Vis/NIR) spectra ranging from 550 to 950 nm were collected in four orientations. Then, calibration models were developed for each orientation separately (local models) and all orientations corporately (global models) to evaluate and compensate for the effect of orientation. After that, the novel method based on the light attenuation theory was introduced to correct the acquired raw spectra and establish corrected orientation models. Results showed that measurement orientation significantly altered spectral intensity due to variations in surface curvature and optical path, thus declining models’ predictive power and robustness. Global models proved to be less susceptible to orientation variation compared with local models. The performance of both local and global models considerably improved post-orientation correction, attributed to the decrease of spectral distribution difference, with their average Rp2 and RPD increased by 93.38 % and 8.11 %, 10.56 % and 10.57 %, respectively, while the average RMSEP decreased by 16.01 % and 10.78 %, respectively. Overall, this work provides a more cost-effective and universal approach to impair the influence of measurement orientation and improve the accuracy and reliability of fruit quality online detection.