Online detection of apples with moldy core using the Vis/NIR full-transmittance spectra

Highlights

• Full transmittance spectra of apple were collected with short integration time mode.

• Effects of orientations and zones on detection accuracy of moldy core was studied.

• The optimal model were built based on LDA classifier coupled with T2 orientation.

• Internal defect was detected using transmittance spectra of infected regions.

Abstract

Moldy core is a common disease of apples, but it is difficult to detect because there is no obvious difference in appearance of fruit. In this study, the full-transmittance spectra of apples were collected online with three different orientations at speed of 0.5 m/s using a short-integration-time mode. Spectral measurement orientation has a great influence on the spectral intensity, but no effect on the spectral trends. The spectral intensity of healthy fruit was higher than diseased fruit for all three orientations due to the stronger absorption of damaged tissues. To detect apples with moldy core, four kinds of classification models including naive bayes (NB), linear discriminant analysis (LDA), extreme learning machine (ELM) and support vector machine (SVM) were developed based on the full-transmittance spectra. The results showed that the spectra extracted from medial zone resulted in better detection performance than for intact fruit, and the T2 orientation was more suitable for moldy core detection. The best classification model was built based on the medial zone spectra collected by T2 orientation with the success rate of 90.4 %, 86.9 % and 93.9 % for total, healthy and diseased samples in the validation set. Overall, it is feasible to online detect moldy core with full-transmittance spectroscopy technology, moreover, the spectral acquisition technology of short-integration-time mode can be used to detect internal defect by extracting the effective discrimination information from infected region.

Introduction

Moldy core (Fig. 1), also known as mould core or core rot, is a common disease affecting the internal quality of apples. Moldy core be generally subdivided into three types including browning, moldy core and rot based on the disease symptoms. Moldy core of apples is easily induced by a variety of fungal spores, such as Fusarium pp., Penicillum sp. and Alternaria alternate sp., and the main pathogenic bacteria vary with the producing area and variety. The confidence of the consumers will be harmed deeply once the moldy core fruit is bitten. Moreover, there are high amounts of mycotoxins in the infected fruit, which exist serious food safety risks for consumers. However, it is almost impossible to prevent moldy core from occurring by spraying fungicide once fungal spore enters to the seed chamber via the calyx sinus after flowering (Ellis and Barrat, 1982; Shtienberg, 2012). Moldy core is undetectable until the fruit is cut or eaten, because the appearance of diseased fruit initially shows no obvious difference with healthy one (Sbenderey et al., 2010).

Moldy core can be visually inspected from the cross-section area after cutting the apple into halves, but is laborious and destructive. Several nondestructive sensing methods have been employed for the internal defect detection of fruit, including X-ray imaging (Herremans et al., 2013; Herremans et al., 2014), magnetic resonance imaging (Gonzalez et al., 2001; Chayaprasert and Stroshine, 2005; Cho et al., 2008) and thermal imaging (Baranowski et al., 2008, 2012). These methods can not meet the industrial inspection requirement for large groups of samples because they are time-consuming. Visible and near infrared (Vis/NIR) spectroscopy has a number of advantages because it is rapid and real-time, with minimal sample preparation. Vis/NIR spectrum could provide useful information about target parameters by measuring the light reemitted from the object in the form of diffuse reflectance and transmittance modes. Diffuse reflectance and transmittance modes are commonly used to acquire the spectra of the tested samples. In diffuse reflectance mode, the spectrograph and lamp are located on the same side of sample, and the reflectance spectra within 800−2500 nm are obtained primarily from superficial regions of fruit (about 1–2 cm thickness) (Fan et al., 2009; Huang et al., 2008). Therefore, the diffuse reflectance mode is mainly applied to detect defects and components of superficial layer (Baranowski et al., 2013; Lorente et al., 2013; Tian et al., 2020; Li et al., 2016). In transmittance mode, the spectrograph and lamp are located on opposite both sides of the tested sample. Thus, the spectrograph can directly acquire the transmission light from the tested fruit and detect the internal properties of fruit. Fu et al. (2007) found that the transmittance mode detected brown heart in pears better than that diffuse reflectance mode. The transmittance mode was more effective as a noninvasive method than reflectance mode for the discrimination of fly infestation in pickling cucumber (Lu and Ariana, 2013). Therefore, the transmission mode is a better choice for the detection of hidden or internal defects in fruit, especially for those of medium size.

Moldy core generally occurs at the core of apples, thus, it is very important to acquire spectral information about diseased tissues to accurately detect the disease. Shenderey et al. (2010) developed an apparatus for detection of moldy core in apples at the speed of one fruit per second based on NIR spectroscopy technology. Hu et al. (2019) developed a classification model of moldy core using the acquired NIR transmittance spectra and back propagation network with a success rate of 95 %. Hence, Vis/NIR transmittance spectroscopy technology accurately distinguished the moldy core, but the lower detection efficiency still limited its practical application. Online detection of moldy core is of great value for apple industry. Recently, a variety of online systems have been developed to detect the internal attributes (Jie et al., 2014; Liu et al., 2014; Tian et al., 2019) and external defects (Huang et al., 2015; Khodabakhshian et al., 2017) of fruit based on Vis/NIR spectroscopy and multispectral imaging technologies.

The objective of this study was to build an optimal classification model for online detection of moldy core in apples. Specific aims were: (1) to analyze the transmittance spectral features collected from three different detection orientations; (2) to investigate the influence of fruit detection orientation on the detection accuracy of moldy core; (3) to determine the most effective spectrum measurement zone for detection of moldy core; (4) to develop an optimal classification model by comparing the performance of different classifiers.