Avocado Ripeness Classification Using a Convolutional Neural Network (CNN)

Abstract

The determination of avocado ripeness plays a crucial role in post-harvest handling and quality The determination of avocado ripeness plays a crucial role in post-harvest handling and quality control within the agricultural sector; however, conventional assessment methods based on visual inspection and human experience are often subjective and inconsistent, potentially leading to classification errors and economic losses. To address this issue, this study proposes an automated avocado ripeness classification system using a Convolutional Neural Network (CNN) based on digital image analysis. The model employs a transfer learning approach using the MobileNet architecture implemented through the Teachable Machine platform. The dataset utilized in this research was obtained from Mendeley Data and consists of avocado images categorized into four ripeness levels: underripe, breaking, ripe, and overripe. Prior to model training, the images underwent preprocessing and data augmentation to improve model robustness and generalization. Model evaluation was conducted using 1,200 test images, with 300 samples per class. Experimental results show that the proposed model achieved an overall accuracy of 91.42%, indicating strong and stable classification performance. Analysis using a confusion matrix reveals that most predictions were correctly classified, while misclassifications primarily occurred between ripeness stages with visually similar characteristics. Among all classes, the underripe category demonstrated the highest performance with minimal classification errors. These findings indicate that the proposed CNN-based approach is effective and reliable, and it has significant potential to be further developed as an automated system for avocado ripeness classification and post-harvest quality assessment.