Mellowness Detection of Dragon Fruit Using Deep Learning Strategy

Abstract

The agriculture being a main source of income in many developing countries such as India, Indonesia, etc. The economic development of these countries depends on the GDP (Gross Domestic Progress) rate of the agricultural products. However due to miscalculations in the maturity of the fruits and vegetables leads to the wastage of foods. In general many measure were taken to minimize the food spoilage and by tracking the each stage of the vegetables and fruits carefully, but resulted in a hefty human labor, and weariness. Specifically the non-climacteric fruit such as the dragon fruit requires much attention as it is has to be harvested after it is ripened and cannot be ripened after harvesting using the hastening ripening process such as the ethylene, carbide, and CO2 etc. So the paper has put forth the application to identify the mellowness in the dragon fruit using the RESNET 152 a deep learning convolution neural network to identify the dragon fruits mellowness and it's time to harvest. The model was trained using the python and the tensor flow. The developed structure was trained using the pictures of the dragon fruit in the different stages of its mellowness and was tested using the region of convergence and the confusion matrix with 100 new data. The testing was carried with the different number of epoch ranging from 10 to 500. The results obtained were more accurate compared to the VGG16 /19 in the terms of Accuracy and loss in training and testing.