Stingless Bee Honey Classification Using Near Infrared Light Coupled With Artificial Neural Network

Abstract

Even though both farm and wild raw honeys are better than processed honey in terms of nutritional value and quality, wild honey is more expensive than farm honey due to its scarcity, nutrition, and quality. However, there is a challenge for consumer to differentiate both farm and wild raw honey due to the complexity of raw honey. Although near infrared (NIR) spectroscopy is promising to assist consumers to differentiate types of honeys, the financial barrier to have a NIR spectroscopy is needed to be addressed. Thus, this research aims to evaluate the performance of a low cost NIR light acquisition alternative in classifying stingless bee honeys using artificial neural network (ANN). First, 164 honey samples of two different types of raw honeys were prepared. Next, NIR light LEDs of five different wavelengths i.e. 850, 860, 870, 890, and 950 nm with light sensors were used to acquire the transmitted NIR absorbance from raw honey sample. ANN with different number of hidden neurons were used to analyze the data, and six datasets were used to investigate the best distance between NIR light source and light sensors. Results indicate that the acquired NIR light coupled with ANN by using eight hidden neurons and an average distance of 40 mm from light source to light sensor were able to produce the best result with the best true positive (TP) correct classification percentage accuracy and cross entropy (CE) value of 96.0% and 1.14, respectively.