Deep learning framework for precision grading and non-invasive Apple sweetness evaluation

Abstract

This research aims to find an economical, non-invasive solution for grading apples by sweetness based on multispectral imaging. This paper examines the relationship between sugar levels and multispectral images of apple fruit taken inside a wooden multispectral imaging enclosure specially designed for taking apple pictures. In addition, using a hand-held refractometer, the sugar content of different apple samples and their corresponding multispectral images are collected. The methodology includes designing and building a prototype camera chamber, gathering pictures of various apples, and using advanced image analysis software to process the data. Prediction of outcomes: Building a practical grading system based on non-invasive multispectral imaging and finding a significant spectral feature of Apple fruit's sweetness levels in the project's range of concern. The proposed system implemented AppleNet, a convolutional neural network (CNN) within MATLAB, to process the multispectral images, and the accuracy achieved for grading the sweetness of apple fruit is 65%.