Artificial Intelligence in Agriculture (Jan 2021)
Mango internal defect detection based on optimal wavelength selection method using NIR spectroscopy
Abstract
A non-destructive technique should be developed for performance analysis of mango fruits because the spongy tissue or internal defects could lower the quality of mango fruit and incur a lack of productivity. In this study, wavelength selection methods were proposed to identify the range of wavelengths for the classification of defected and healthy mango fruits. Feature selection methods were adopted here to achieve a significant selection of wavelengths. To measure the goodness of the model, the dataset was collected using the NIR (Near Infrared) spectroscopy with wavelength ranging from 673 nm–1900 nm. The classification was performed using Euclidean distance measure both in the original feature space and in FLD (Fisher's Linear Discriminant) transformed space. The experimental results showed that the lower range wavelength (673 nm–1100 nm) was the efficient wavelength for the detection of internal defects in mangoes. Further to express the effectiveness of the model, different feature selection techniques were investigated and found that the Fisher's criterion based technique appeared to be the best method for effective wavelength selection useful for classification of defected and healthy mango fruits. The optimal wavelengths were found in the range of 702.72 nm to 752.34 nm using Fisher's criterion with a classification accuracy of 84.5%. This study showed that NIR system is a useful technology for the automatic mango fruit assessment which has the potential to be used for internal defects in online sorting, easily distinguishable by those who do not meet minimum quality requirements.
