Thermal Imaging Based Defect Detection in Pomegranate with Machine Learning and Deep Learning Techniques

Abstract

Fruits and vegetables are perishable goods and undergoes deterioration right from post-harvest. Fruit quality is determined by both internal and external factors, encompassing a wide range of physical, chemical and biological attributes that affect the appearance, taste, nutritional value, shelf life and market appeal of fruits. The internal factors are inherent characteristics are primarily governed by taste, aroma, texture, ripeness and nutritional composition whereas, the external factors of appearance, visual defects, size and shape. For evaluating the quality, internal as well as external characteristics need to be taken into account. The external characteristics can be assessed through RGB image processing techniques. The idea of internal characteristics can be obtained from the thermal image. In this paper, a machine learning and deep learning approaches are proposed for quality inspection of pomegranate with thermal imaging. A machine learning models are trained with KNN, SVM, Random Forest and CNN techniques for defect detection and classification of pomegranates in the categories of no defect, minor defect and major defect. The CNN approach outperforms with 92.78% accuracy as compared to 79.26% of SVM and 84.07% of random forest.