An Automated Framework for Mango Fruit Processing using Computer Vision Techniques

Abstract

In the era of advanced artificial intelligence, the development of vision-based models for fully automated manufacturing processes in the fruit industries is of paramount importance. Adapting to intelligent systems to carry out manufacturing in fruit processing industries ensures the sustainability and quality of end products. This aspect of the intelligent manufacturing process is especially true for Alphonso fruit processing. India is one of the leading countries concerning mango fruit-based products. Multiple challenges must be meticulously addressed to fulfill the growing demand for mango fruit and maintain high standards of fruit quality and precision. newlineThis thesis presents a set of advanced vision-based algorithmic models for automated Alphonso processing with a specific focus on precision weight estimation, ripening stage classification, and defect detection. The proposed contributions employ the concepts of image analysis, computer vision, machine learning, and deep learning to address the aspects involved in fruit grading and sorting in industrial manufacturing. The present investigations propose enhancing existing computational models and contributing to the comprehensive datasets for Alphonso mangoes, precision weight estimation, and maturity classification by adapting to natural ripening methods.The initial dataset contributions developed for the study include AMDPWE (Alphonso Dataset for Precision Weight Estimation) for vision-based weight estimation. AMDPWE includes the sample images of the Alphonso cultivar with multiple orientations. The other dataset contributions include AMDRSC (Alphonso Dataset for Ripening Stage Classification), comprising of naturally ripened mango image samples capturing the day-to-day textural variations in six orientations, and AMDDD (Alphonso defect Detection Dataset) consisting of defected fruit with bruises, bacterial black spots and rotten. The datasets AMDPWE, AMDRSC, and AMDDD serve as the foundation for the proposed computational model contributions on vision-based..