Efficient Fruit Grading System Using Spectrophotometry and Machine Learning Approaches

• Published in: IEEE sensors journal Vol. 21; no. 14; pp. 16162 - 16169
• Main Authors: Chopra, Hetarth, Singh, Harsh, Bamrah, Manpreet Singh, Mahbubani, Falesh, Verma, Ashish, Hooda, Nishtha, Rana, Prashant Singh, Singla, Rohit Kumar, Singh, Anant Kumar
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.07.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Artificial intelligence; Cameras; Cloud computing; Computer vision; driverlessAI; fruit grading; Fruits; Grading machines; Human error; Industrial plants; Machine learning; Model accuracy; Model testing; Python; Sensors; Spectrophotometry; Spectroscopy; Training
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2021.3075465

Physical Classification of ripe fruits is an expensive affair in the agriculture industry and human error can lead to inaccurate results. This paper introduces the concept of an intelligent AI-based system using spectrophotometry and computer vision for automated fruit segregation based on their grade. When the fruit is fed into the proposed system, the fruit is identified with 95% accuracy, using a cloud-computing platform provided by Microsoft Azure. After that, using spectroscopy and ensemble machine learning approaches, fruit grade is predicted. This ensemble model is trained using 1366 apple readings taken from Unitec's Apple Sorting and Grading Machine from an industrial plant. With the help of H2O's Driverless.AI, the proposed ensemble provides an overall approximate validation accuracy of 82%. The model is also tested on an unseen test dataset containing real-life spectral values and the accuracy of fruit segregation into different classes peaked at 72%.