Abstract

Bitter pit is a physiological disorder in apples. Several major apple varieties are susceptible to this disorder that poses a great challenge to growers and the associated industry as it significantly reduces the produce utilization value and marketability. The current method of bitter pit detection is through visual assessment of symptoms. Near infrared (NIR) spectroscopy is a non-invasive technique that can be utilized towards detecting bitter pit development in fruits in pre-/non-symptomatic stages. Therefore, NIR spectra (935- 2500 nm) of apples were collected from healthy and bitter pit Honeycrisp, Golden Delicious, and Granny Smith apples from a commercial orchard. The apples were stored in a controlled environment and spectral reflectance data were acquired at days 0, 7, 14, 35, and 63 after harvest. Chemical analysis was performed at the end of the storage period to estimate calcium, magnesium, and potassium content in the fruit peel. Partial least square regression (PLSR) was used to identify the apples as healthy or bitter pit using NIR-based spectral features. In addition, specific spectral features were selected by implementing two feature extraction methods: PLSR and step-wise discriminant analysis (SDA) on day 63 spectral data-set. The PLSR and SDA-based selected features from day 63 in Honeycrisp apples classified the same dataset with classification accuracies of about 100% with both methods. Regression analysis indicated a strong relationship between the PLSR-based spectral features and magnesium-to-calcium ratio in fruit peel in all three ( Honeycrisp, Golden Delicious, and Granny Smith) apple varieties.