Assessing Dragon Fruit Pulp Health Through Machine Learning and Deep Learning

Abstract

Dragon fruit is having high nutritional value and health benefits and it is increasingly popular worldwide. This fruit is rich in antioxidants, minerals, and vitamins, supports immunity, digestion, and skin health. Dragon fruit, though nutritionally beneficial, is susceptible to various diseases that affect its yield and quality. In India, around 2-5% of dragon fruit farmers have reported the fruit defect which causing a formation of white structure within a dragon fruit pulp, leading to concerns over fruit health and market value. The formation of white structure can be termed as “meshy” pulp formation in the fruit.

This research focuses on detecting the meshy structure within dragon fruit pulp, which indicates poor fruit health. 120 fruits were sampled, cut, and imaged to create a dataset for analysis. To classify pulp health based on image data, a hybrid approach is developed with deep learning and machine learning techniques. Specifically, we utilized feature extraction layers from two deep learning models-Inception and ResNet. These models, known for their accuracy in image feature detection and gives intricate details in the pulp texture. The extracted features from the Inception and ResNet models were integrated with a machine learning-based classification layer to identify meshy versus non-meshy pulp.

This hybrid model provides an effective solution for detecting presence of the meshy structure from images, facilitating a quick and reliable health assessment of the fruit. This research offers a valuable solution for the dragon fruit processing industry by enabling automated identification of pulp health, ultimately enhancing product standards and reducing the need for manual inspection.