Estimating Mango Mass from RGB Image with Convolutional Neural Network

• Published in: 2022 3rd International Conference on Artificial Intelligence and Data Sciences: Championing Innovations in Artificial Intelligence and Data Sciences for Sustainable Future, AiDAS 2022 - Proceedings
• Main Author: 2-s2.0-85141775201
• Format: Conference paper
• Language: English
• Published: Institute of Electrical and Electronics Engineers Inc. 2022
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85141775201&doi=10.1109%2fAiDAS56890.2022.9918807&partnerID=40&md5=c0881a97c21b4e77af68bd4846a5053f

Mangoes are graded based on their textures, colors, sizes and masses. Thus, there are large number of studies focussing on the size and weight estimation of mangoes. Among the methods that have been explored for mango mass estimations are SVM, discriminant analysis, pixel counting, Bayesian networks, and ANN. However, there are very few studies that use Convolutional Neural Network to estimate mango mass. Consequently, this study presents a CNN architecture capable of accurately estimating the mass of Harumanis mangoes from RGB images. The primary contributions of this study are CNN architectures capable of estimating the mass of a Harumanis mango using only computer vision and a blank A4 paper as visual cue and Harumanis mango image and mass dataset for benchmarking. The CNN architectures were trained on 548 sample images of mangoes that were into training and validation sets. The results have been promising as the CNN architectures are capable of producing mango mass estimation (in kilograms) with MSE values of 0.00227 and MAE values of 0.03697, which are close to the actual mass. © 2022 IEEE.