Multi-Parameter Based Mango Grading Using Image Processing and Machine Learning Techniques

Abstract

Mangoes, the king of fruits is globally exported and locally consumed on a large scale. During exports and local marketing, delivering good quality fruits and satisfying certain pre-defined standards is important. This post-harvest operation of quality checking, known as mango grading is usually performed manually. But manual grading can be in-consistent, erroneous and labor-intensive. A computer vision based grading solution will result in consistent and accurate sorting. Such a mango grading system based on external parameters namely ripeness, size, shape, defects was developed in this research work. Image processing techniques were applied to extract the color, geometric and shape related features. These features were further utilized by pre-trained random forest classifiers to determine the mango ripeness (unripe/mid-ripe/ripe), size (small/medium/large) and shape (well-formed/deformed) category. K-means clustering was applied for defect segmentation to determine the mango defect category as (non-defective/mid-defective/completely-defective). Final grading was performed using a grading formula that combines the parameter specific quality scores assigned, according to predicted categories. Ripeness, size and shape classification performed on a created dataset of Dashehari mangoes achieved a test accuracy of 100%, 98.19% and 99.20% respectively. Formula based integrated grading could grade mangoes with 88.88% accuracy.