Dataset creation of thermal images of pomegranate for internal defect detection

Datasets are crucial in various fields, especially in the context of machine learning, data science and research. Datasets are used to train machine learning models. A model learns patterns and relationships from the data it is exposed to. The dataset used for training a machine learning model shall be diversified and consist sufficient samples of desired categories. This paper presents various steps and its outcome in preparing the dataset of digital and thermal images of pomegranate for recognising internal defects. The defects in fruits are often categorised as surface defects and internal defects. The surface defects are recognised with digital RGB image but fails to give insight about the internal structure of the fruit in which we are often interested. The thermal images can be used to detect the internal defects in fruits. When a fruit is subjected to temperature difference as compared to the surrounding, the thermal emissions from fruit captured through a thermal camera (thermal image) gives the key information about the internal damages in the fruit. The internal defects are reflected in thermal image as variations in temperature of adjacent pixels. The k-mean segmentation is applied for identifying internal defects with thermal images in pomegranates to categorize them viz. No defect, major defect and minor defect. This information is useful for training a machine learning algorithms that are intended for bulk processing in the field of fruit defect detection and classification.