Maturity prediction of banana with different bunch cover materials using artificial neural networks

Abstract

Banana production in the northern part of Thailand has grown rapidly in the past few years. Bunch cover is the common practice to prevent disease infection, damage of insects and thermal damage. Generally, the commercial bunch cover comprises several layers of the material, i.e., thin non-woven, paper, foam, and plastic sheet. Unfortunately, after harvesting, the excessive amount of the bunch cover materials cannot be reused and is always dumped as a landfill which might lead to environmental problems soon. In order to reduce waste, the recyclable non-woven based material (Tyvek) was proposed to be used as a new bunch cover material. The experiment was conducted at the banana orchard in Phaya Meng Rai District, Chiang Rai Province during December 2016-February 2017. Bananas were harvested at 15 weeks after inflorescence emergence. Image analysis was employed to analyze banana qualities, i.e., color and fruit roundness. The results showed that the temperature inside the bag played an important role for fruit maturity. The fruit weight, length, pulp/peel ratio and fruit roundness of the commercial bunch cover were significantly greater than those of Tyvek bunch cover (P>0.05). As for the lightness of banana, the Tyvek bunch cover provided a significantly darker peel color (P<0.05) due to a greater amount light penetration intensity. In terms of maturity prediction, the predicting models were developed using artificial neural network. Temperature, bunch cover type, light intensity and hand location were assigned to be the inputs. On the other hands, hand weight, color, fruit length and peel/pulp ratio and fruit roundness were used as the outputs. The different model architectures, number of nodes, and learning functions were tested. The results showed that the developed 'Wardnet' with 3-hidden-layer provided the best prediction results with highest R2 (0.97-0.98) and with lowest SE (0.01-0.04).