Abstract:
Near-infrared (NIR) spectroscopy is growing as a valuable non-destructive analytical technique,
acquiring prominence in a variety of application areas. This thesis describes the creation of a lowcost,
portable NIR-based device for classifying fruits and grading milk. Using NIR sensors
AS7263, the device measures the absorption of light by fruit samples at six distinct wavelengths.
The data gathered by the aforementioned sensors is processed by machine learning algorithms in
order to classify produce and grade milk. The device includes a web application that enables realtime
viewing of classification results, thereby facilitating the making of informed decisions. The
results of principal component analysis (PCA) indicate that the proposed device can effectively
differentiate between various fruits and accurately detect variable milk-water percentages. Two
distinct R-based machine learning models demonstrate that the classif.ranger algorithm can
effectively classify fruits based on their spectral reflectance values and milk samples based on their
spectral reflectance data with an accuracy of 0.964% and 96.36%, respectively. The analysis
verifies that machine learning algorithms can be utilized effectively in the agriculture and dairy
industries for quality control and food safety applications. The proposed device provides a costeffective
alternative to the costly spectrometers presently on the market, which can be prohibitive
for researchers and producers.
