A Comparison of Deep Learning Based Methods for Grading Papaya Fruit by Maturity Level

Papaya is a tropical fruit with high nutritional value. Identifying the maturity level of papaya is important in fruit industries. However, doing this manually by humans is inefficient. To automate this process, we tested two deep-learning models for classifying papaya images into four categories: ripe, partially ripe, unripe, and decayed. Previous works are focused on the classification of papaya fruit maturity into unmatured (unripe), partially mature (partially ripe), and mature(ripe), we have extended this to one more class i.e. Decayed. This class inclusion plays a vital role in distinguishing the decayed fruits from healthy ones in real-world applications and also helps in preventing the distribution of compromised produce, thereby maintaining quality control in the food supply. The primary goal is to find the best-performing deep learning model for accurate classification, which is essential for applications in agriculture and the food industry for quality control and automated sorting. We tested well-known Deep learning models: DenseNet121 and InceptionV3. Both the trained models are evaluated on the dataset comprising 1809 images. Each model's performance is evaluated using a different validation set and its classification accuracy. Our experimental results shows that the InceptionV3 model performs with the highest accuracy at 99.37% at 90:10 train-test ratio with batch size 16 under 40 Epochs, demonstrating superior performance in distinguishing between the four categories of papaya images.