‘Fuji’ apples produced in four counties of Shaanxi Province, China were used as samples to explore potential applications of portable spectrometers in determining sugar content and firmness of apples produced in different areas. Eighteen and twenty wavelengths were selected as characteristic wavelengths (CWs) using successive projections algorithm (SPA) from pretreated full spectra for sugar content and firmness, respectively. Two linear models (multiple linear regression (MLR) and partial least squares regression (PLSR)) and other two nonlinear models (general regression neural network (GRNN) and extreme learning machine (ELM) were adopted to build sugar content and firmness determination models. The results indicate that not only for sugar content, but also for firmness, PLSR had better performance than MLR, and ELM performed better than GRNN. PLSR‐SPA had the best determination performance for sugar content and firmness. The research offers useful spectrometer technologies on developing portable detectors for internal qualities of apples. Practical applications ‘Fuji’ apples are very popular among consumers around the world since they are sweeter and crisper than many other apple cultivars. Although several studies have been conducted on predicting internal qualities of ‘Fuji’ apples by using NIR spectroscopy, few studies have been reported on sugar content and firmness determination of ‘Fuji’ apples produced in different areas. In this study, a portable near‐infrared spectrometer and diffuse transmittance spectroscopy was used to determine sugar content and firmness of ‘Fuji’ apples produced in different areas of Shaanxi Province, whose ‘Fuji’ apple production and plant area ranks the first in China. This study offers useful technologies for developing portable detector and online detection instrument on quality attributes of ‘Fuji’ apples.

No abstract