3. Results and discussion
3.1. Investigation of the instrument setup for optimal spectral acquisition
Due to each mangosteen sample having different internal factors such as translucent flesh, hardening pericarp and seeds, the dynamic ranges of the SW-NIR intensity spectra were very different, especially, between translucent flesh and hardening pericarp. The hardening pericarp, which is usually produced after a physical shock during harvesting and transporting, caused low signal-to-noise ratio spectra because its tissue has low light-penetration. Therefore, preliminary tests were conducted to determine the optimum conditions for measuring all mangosteens. About 1000 mangosteens were thus used for the investigation of the measurement conditions using the basic procedure described above.
3.1.1. Investigation of light sources and SW-NIR spectra
Different light source levels of 100, 200 and 400 W were tested for this experiment. The SW-NIR spectra of mangosteens featured noise and a low-level intensity when using a low-level light source compared to using a higher level light source. The spectra intensity of translucent flesh was excessive when using a 400 W light source. However, at 100 W mangosteens with hardening pericarp yielded very low-level spectra and signal-to-noise ratios while the spectra of mangosteens with translucent flesh were more strongly produced.
The results showed that the light level is one of the important measurement conditions in order to correctly obtain spectra of translucent flesh and internal disorders of mangosteens. Therefore, we decided to use a 200 W light source, which was deemed appropriate for this study. This was to avoid excessive intensity with translucent flesh and to have high signal-to-noise ratio spectra with hardening pericarp mangosteen.
3.1.2. Investigation of measurement speed effect on SW-NIR spectra
The SW-NIR spectra of samples were collected at different measurement speeds (3, 6 and 12 m/min). We found that the spectra of mangosteen samples were better at the lower speeds in terms of the signal-to-noise ratio. Therefore, the optimum measurement speed for this study was selected at 3 m/min with an integration time of 78 ms.
3.1.3. Investigation of the influence of fruit holder on SW-NIR spectra
A strong light source is necessary for mangosteens because of their thick pericarp and low penetration characteristics. However, the light could get to the detector directly at the contact surface area between the fruit body and silicone sample holder. As a result, sample holders were designed to have a shield of silicone rubber on top to support the fruit body in order to eliminate the clearance at such contact surface areas. One, two and three shields of silicone rubber were arbitrarily tested. The results showed that the use of two or three shields was better because the SW-NIR spectra were not influenced by direct light sources. Two shields of silicone rubber were therefore used throughout the main experiment.
3.2. Fruit classification
3.2.1. Effect of fruit size and seed on the SW-NIR spectra of intact mangosteen
The averaged SW-NIR intensity spectra of intact normal mangosteens and translucent flesh mangosteens with variation in fruit size are shown in Fig. 3(a). Fruit size had an effect on the SW-NIR intensity spectra, but the difference in the intensity in some regions resulting from the translucent flesh was substantially greater. Fig. 3(b) shows the effect of seed on the averaged SW-NIR intensity spectra. Similarly, translucent flesh with or without seeds caused much greater intensity than normal flesh.