Effective modification through transmission Vis/NIR spectra affected by fruit size to improve the prediction of moldy apple core

Highlights

• A method to correct the effect of fruit size on the NIR spectra was proposed.

• The method was to facilitate the identification of internal diseases in apples.

• An equation to calculate extinction coefficient of transmitted light was developed.

• The extinction coefficient was used to modify the transmission spectrum.

Abstract

Moldy core is an internal disease of apples that can be detected using transmission near-infrared spectroscopy. However, the transmission spectra of apples are affected by the size of the fruit (i.e., the optical path length). When transmission spectra are used to distinguish between healthy and diseased apples, healthy apples with large diameters are often misclassified as diseased, while apples with moldy cores and small diameters are often misclassified as healthy. To solve this problem, a spectral correction method, based on the size of the fruit is proposed herein. Spectra and transverse diameter information for 327 Fuji apples were obtained in this study. Using experiments, we verified that the rate of light extinction in apple flesh varied log-linearly with thickness. In this case, spectra of healthy apples with diameters of 80 mm were selected as a reference. Comparing the spectra of 327 apples with varying diameters against this reference, we developed an equation to calculate the extinction coefficient of transmitted light within the fruit. Transmission spectra were modified according to this extinction coefficient. Error back propagation artificial neural networks and support vector machine models were established based on corrected versus original spectra. These results showed that the accuracy of the classification models based on the corrected spectra was much higher, yielding accuracy rates of 98.04% for the training set and 90.20% for the test set. Clearly, the effect of fruit size on the transmission spectra could be corrected to improve the identification of diseased apples using this method.

Introduction

The apple is a fruit that is appreciated around the world because of its texture, flavor, visual appeal, and nutritive values [1]. Because of the ongoing improvement in people’s living standards, the demand for quality apples is also increasing [2]. However, fungi often infect apples because of the acidic pH of their tissue, causing them to develop pathological disorders [3], [4]. Moldy core is one of the most common pathological disorders, which is characterized by dry, spongy brown lesions in the core area. In some cases, a moldy core can turn into core rot, in which the infection extends into the fruit flesh. A series of fungi, such as Penicillum expansum, Mucor piriformis, and Alternaria alternate, can enter the fruit through an opening in the calyx and attack the core region and adjacent flesh [5], [6], [7]. There is no difference in the external characteristics of apples with moldy cores and healthy ones; thus, the moldy core is undetectable until the fruit is cut open or bitten into. If such apples enter the market, they can pose problems to both producers and consumers.

To detect such internal defects in apples, past practices required a destructive test in which apples were cut into halves and the cross-section area of each half was visually inspected [8]. However, this method is tedious and laborious, as well as destructive. More recently, several nondestructive testing methods to detect internal defects in apples have been employed, including X-ray imaging [9], [10], magnetic resonance imaging [11], [12], [13], [14] and thermal imaging [15], but applications of these methods are typically expensive. Near-infrared transmission spectroscopy (NIRS) is a nondestructive testing technology that has been widely used for internal disease detection in apples because of its fast analysis speed and low cost. This technology has been developed since the 1980 s and is now largely mature [16], [17], [18], [19], [20], [21], [22], [23], [24].

The capability of detecting internal disorders in apples using NIRS was demonstrated by Upchurch et al. [25]. A classifier based on the ratio between the light intensity at 720 and 810 nm was used to segregate apples with internal breakdown from good apples. In this research, ‘Delicious’ apples were stored at 0℃ for 6 months, providing ample time for the development of internal breakdown. Meanwhile, the effects of the source-detector arrangement and fruit orientation on browning detection in ‘Braeburn’ apples were investigated by Clark et al. [26]. To address detection errors caused by the uneven distribution of brown tissue, these authors suggested using multiple measurements from different positions. They also pointed out that the various sizes of the apples introduce classification errors. However, although they mentioned that common normalization methods, such as multiplicative scatter correction (MSC) or standard normal variate (SNV), should be appropriate for dealing with variations in fruit size, they did not explore ways to nullify the effect of fruit size on transmission measurements of apples in any depth. To non-destructively measure the percentage of internal tissue browning in ‘Braeburn’ apples, McGlone et al. [27] developed two prototype on-line NIRS systems. One system was based on a time-delayed integration spectrometer and the other was based on a large aperture spectrometer. The latter system had superior accuracy, yielding a root mean square error of prediction of 4.1% after partial least squares calibration. However, only apples with a mean equatorial diameter of 76 mm (SD = 2.8 mm) were used, and the effect of different fruit sizes on detecting browning tissue was not studied. Shenderey et al. [28] developed a rotating system, equipped with an off-the-shelf miniature spectrometer to simulate on-line detection of moldy cores in apples. The accuracy of their classification results was high, with 92% detection of healthy apples and 100% detection of decay at levels above 30%. Despite all of these studies achieving successful detection of internal diseases in apples using NIRS, none have considered the effect of fruit size on the transmission spectrum, nor did they correct the spectrum to address this effect. This has led to some misidentification of healthy apples and mildly infected apples. Herein, we used Vis/NIRS to detect moldy cores in apples and analyzed all misclassified apples. We found that all examples that were wrongly classified were either healthy apples with large diameters or diseased apples with small diameters. We found that correcting the transmission spectrum based on the size of the apple reduced such discriminant errors in classifying apples.

The NIR spectrum measured at the equator of the fruit shifts upwards, as the diameter of the fruit increases, as reported by Kawano et al. [29]. They suggested that the effect of fruit size can be reduced by dividing the 2nd derivative spectra by the diameter of the fruit. However, this method is not convenient for routine applications. Instead, a wavelength with a 2nd derivative value having a high correlation to the diameter of the fruit was selected as a correction factor for fruit size. Using this correction factor, the corrected spectra accurately estimated the Brix value of satsuma mandarins. Further study found that the rate of light extinction in apple flesh is log-linear with thickness [30]. This finding was confirmed in our experiments. The transmission of light in apples was quantified by Qin and Lu [31], [32] using Monte Carlo simulations. To eliminate any distortion of light intensity caused by the curved fruit surface, they corrected the diffuse reflectance spectrum based on the size of the fruit.

Based on the above research, this paper proposes a new method to correct for the effect of fruit size on the transmission spectra of apples, facilitating identification of internal diseases in different sized apples.