Intelligent Apple Fruit Quality Grading System Using Deep Learning

With the significance fruit plays in overall human health, the global demand for fresh produce grows every day. However, fruit selection that is dependent on human evaluation remains time-consuming, prone to inconsistencies, and waste-inducing. Therefore, automating this process using a fruit classification system could provide a cost-effective solution that aligns with the Sustainable Development Goals (SDGs). This paper proposed an intelligent Apple quality grading system. The proposed system consists of four main phases; preprocessing, classification, post-processing, and analytical report. In preprocessing, data oversampling followed by data augmentation techniques are applied. Then, the processed data is used to feed the MobileNetV2. Then, in the post-processing phase, the Segment Anything Model (SAM) is utilized to extract the rotten part from the apple fruit image. Then, based on the calculated standard deviation and specified threshold value, the quality of the apple fruit is classified as either high or low. Finally, the classification result will be reported in the report with the calculated fresh-to-rotten fruit percentage. The performance of the proposed system is evaluated and tested on two datasets; a private dataset and a benchmark dataset. The results revealed that MobileNet V2 obtained the best results compared to other deep-learning architectures. It obtained an accuracy of 96.67% and a F1-score of 96.84% respectively, on the private dataset. Comparing state-of-the-art models, the proposed system obtained an accuracy of 99% using the benchmark dataset.