IoT-Enabled Modified XGBoost Approach for Ripeness Detection and Classification of Bananas in Smart Agriculture

Abstract

   The ripening stage determination for the climacteric fruit banana bears great importance in terms of its  medicinal and food values with good commercialisation. The ripening of banana fruit incurs huge loss specially in case of transit , shipping  and storage . The large-scale handling  of the  fruits leads to the bulk loss. The biochemical process is the effective one for such determination. Most of such procedures are invasive, intrusive, harmful and create some false interpretation due to insufficiency of illumination during different times of day. In the present study the texture features through GLCM  corresponding to different phases of ripening in banana species has been examined. The IOT sensor interpreted aromatic data difference, its classification and prediction of rotting has been carried out. The modified XGBoost algorithm optimised by the modified Grid search algorithm has been implemented where the enumeration of the split points based on the 1^st and 2^nd order gradient has been carried out. On the basis of split score value the gain has been estimated and the child on left and right has been assigned. For the grid search model, the latin hypercube concept has been used. Feature vectors of approximately 2240 different samples have been prepared in training the model with the achievement of accuracy value of 98%. The classification result for the banana cultivars Martaman, Singapuri and Kathali in the present study has been compared to mathematical models based on ripening characteristics along with other traditional models. The result shows  considerable improvement confirming  potential aspects of  the proposed model towards  smart farming.