In-Field Assessment of Olive Fruit Quality Using a Low-Cost Multispectral Sensor and ANN Models

Abstract

Optimizing harvest time and oil production requires accurate olive fruit quality characterization. Traditional chemical methods are costly and tedious, leading to poor monitoring resolution and reliance on subjective visual assessments. While spectroscopy offers a non-destructive alternative, standard equipment remains complex and prohibitively expensive for smallholder farmers. To address this, we propose a methodology using a custom-made, low-cost multispectral device. Built upon the AS7265x board, the system acquires 18 spectral bands in the visible and near-infrared range (410–940 nm). We used these spectral data to feed artificial neural network (ANN) models for estimating the quality of intact olives. During a two-season field experiment, we monitored ripening to acquire spectral signatures and ground-truth values for oil content per fresh weight (OCFW), oil content per dry matter (OCDM), moisture (M), and titratable acidity (TA). External validation showed high accuracy for OCFW (R2p = 0.86), OCDM (R2p = 0.86), and M (R2p = 0.89), proving the system’s reliability. However, TA estimation showed lower performance (R2p = 0.21), indicating limited spectral correlation. These findings pave the way for affordable, real-time smart farming tools for olive quality monitoring.