Evaluation of a handheld near-infrared spectroscopy sensor for rapid corn kernel moisture estimation

Abstract

Frequent monitoring and accurate estimation of corn (Zea mays L.) kernel moisture are necessary for timing harvests and maximizing profits. Harvesting grain above the U.S. market threshold (15.5%) increases the risk of grain shrinkage and cost of artificial drying, and leads to a loss in profitability (Martinez-Feria et al., 2019) as well as grain quality concerns (Chai et al., 2017). Among corn growers, the standard ways to estimate kernel moisture involve tabletop and portable grain analysis computers (GAC; Sadaka & Rosentrater, 2019). However, GACs require destructive ear sampling and regular calibrations which can be time and labor intensive, so growers might only collect ears from small areas of large fields. Due to spatial variation in grain moisture in large fields (Miao et al., 2006), moisture estimated by sampling ears from a small area will often be unrepresentative of field-level kernel moisture content. Therefore, there is an urgent need for handheld sensors that can accurately and rapidly estimate kernel moisture to conduct sampling over larger spaces in a timely and cost-effective manner.

Commercially available pin and ring-type handheld sensors that can conveniently and rapidly estimate kernel moisture content non-destructively have been evaluated (Fan et al., 2020; Fan et al., 2021). Although these sensors have shown high accuracy (R² = .80), there are some significant concerns. For example, pin-type sensors poke holes into kernels, resulting in severe kernel damage (Fan et al., 2020). Ring-type sensors, on the other hand, operate only within a narrow kernel moisture range (15–32%) and specific ear diameter (2 inches) (Fan et al., 2020). It is essential to evaluate new handheld sensors that can potentially estimate a wide range of kernel moisture content across ear sizes without damaging the kernels. Therefore, the objective of this study was to evaluate the accuracy of a handheld sensor that operates using near-infrared spectroscopy (NIRS) technology to estimate kernel moisture content in-field.

The NIRS sensor evaluated, commercially known as SCiO kernel moisture analyzer (Consumer Physics, Israel), has three major components: the SCiO sensor, the mobile app to operate the system, and the cloud-based telemetry system (Figure 1). The wavelength and ambient temperature range for sensor operation is 750–1050 nm and 32–100 ^οF, respectively. To measure kernel moisture, SCiO is fixed to an oval-shaped corn sensor accessory (2.7 inches long and 1.9 inches wide), attached to the dehusked ear (Figure 2a), and scanned five times. For each scan, the kernels are illuminated with near-infrared light. The reflected spectrum is measured and sent to the cloud, where chemometric models and machine learning algorithms are used to estimate moisture based on past observations and developmental data. Finally, the average kernel moisture is displayed on the mobile app (Figure 2b). On average, only 20–30 s are needed for scanning each ear.

The study involved measuring kernel moisture on ears sampled from nine corn hybrids (ranging from 85 to 109 comparative relative maturity) planted across multiple dates (between April 28 and June 15) in Lansing, MI, over three growing seasons (2020, 2021, and 2022). To capture a wide corn kernel moisture range, we ensured that an appreciable number of samples was taken from dry ears (moisture < 20%), medium-dry ears (between 20 and 35% moisture), and wet ears (moisture > 35%). A total of 70, 87, and 43 ears were sampled in 2020, 2021, and 2022, respectively. In the field, individual ears were dehusked, and kernel moisture was measured mid-ear by placing the NIRS sensor on the upward-facing side of the ear (Figure 2a). After scanning each ear, kernels only within the scanned area (2.7 inches by 1.9 inches) were carefully removed, their total fresh weight measured, and then placed in coin envelopes. Samples were then placed in the oven at 104^οC for 72 hours to measure dry weight, and wet and dry weights were used to calculate moisture content for each sample.

Data were analyzed separately for each year by fitting a linear regression using the “lm” package in R (Bates et al., 2015). In the linear model, oven-dried moisture was considered the independent variable, while NIRS sensor moisture was considered the dependent variable.

Kernel moisture content of the samples ranged from 13 to 70%, 15 to 60%, and 15 to 55% in 2020, 2021, and 2022 respectively (Figure 3). There was a significant correlation (p < .001) between the NIRS sensor and oven-dried moisture content with a high coefficient of determination (R² ≥ .95) across all the years. Predictive accuracy of the NIRS sensor was also high across wide moisture ranges (∼15–60%) studied in each of the three years, indicating its potential use across a variety of corn end uses including high moisture corn (harvested at 28–34% moisture) and earlage (harvested at 35–40%).

The NIRS sensor performance was similar to other sensors evaluated in previous studies (Fan et al., 2020, 2021) but with higher predictive accuracy. The higher accuracy could be due to multiple reasons. First, the NIRS model has better predictive accuracy than resistor-capacitor impedance models used in other sensors (Esteve Agelet & Hurburgh, 2014). Second, kernel moisture is estimated from multiple scans (≥5) and hence reduces any potential variability. Moreover, the NIRS sensor has a large area (4 inch²) occupying up to seven kernel rows. As there are typically 16–18 rows of kernels per corn ear, the NIRS sensor estimated moisture from around 40% of the total rows of kernels and may help explain the relatively higher precision.

The NIRS sensor showed strong potential in accurately and rapidly estimating kernel moisture in the field across a wide moisture range (15–60%). Sensor accuracy was also consistent across multiple corn hybrids and environments evaluated in this research. For corn growers, timely and accurate estimation of kernel moisture is helpful in facilitating optimal harvest timing. Our results showed that the NIRS sensor could be a suitable option for rapid and accurate estimation of kernel moisture in the field. Future studies could explore the feasibility of using this sensor to estimate kernel dry down in the field and its relation to environmental conditions.

B. Agyei: Data curation; formal analysis; investigation; visualization; writing—original draft. J. Andresen: Conceptualization; methodology; supervision; writing—review & editing. M. P. Singh: Conceptualization; funding acquisition; investigation; methodology; project administration; resources; supervision; visualization; writing—review & editing.

The authors declare no conflicts of interest.