Automated Grading of Citrus suhuiensis Fruit Using Deep Learning Method

Abstract

An automated grading system is important in assisting the farmers to perform quality inspection in a more effective manner as compared to manual approach. Besides that systematic fruit grading is a requirement for effective fruit and vegetable marketing. This is because delivering immature, and bruised fruits will lead to lower market price. Hence, this work proposed an automated Citrus suhuiensis fruit grading system based on image processing that can detect multi-index simultaneously such as maturity, quality and size of a local fruit. The fruits are classified according to the grading specification provided by Federal Agricultural Marketing Authority (FAMA). A convolutional neural network method is adopted to perform the classification process. A total of 303 training images and 75 test images were used in maturity dataset, whilst total of 283 training images and 68 test images were used in quality dataset. Experimental results showed that the proposed classification model able to classify the fruits into 6 classes of maturity and 3 classes of quality.