IDENTIFYING THE RIPENESS AND QUALITY LEVEL OF STRAWBERRIES USING DEEP LEARNING

Abstract

Strawberries are one of the most popular fruits in Indonesia. In 2022, the Central Statistics Agency (BPS) recorded strawberry production at 28,895 tons, a 193.05% increase from the previous year. West Java was the main producer, with 25,413 tons, accounting for 87.95% of total production. However, the sorting process is still done conventionally to determine the ripeness and quality of the fruit, which consumes time and resources. This system utilizes deep learning technology with YOLOv7 and EfficientNetV2S models, integrated with the cloud and implemented on an Android application. The app sends images to the deep learning system, which processes and classifies the ripeness of strawberries. The smartphone displays the confidence level and classification results. Based on testing, the system can identify five categories: Ripe Grade-A, Ripe Grade-B, Half-Ripe Grade-A, Half-Ripe Grade-B, and Unripe. The loss values for Box and Val Box are 0,02095 and 0,03029, respectively; Objectness and Val Objectness are 0,004057 and 0,00333; Classification and Val Classification are 0,008343 and 0,007392. The classification model evaluation showed precision, recall, and F1-Score of 0.990 each and an accuracy of 99%. Cloud processing time reached 1-2 seconds with object classification at 180 milliseconds. Usability testing with 33 respondents showed dominant scores of 4 and 5, and the application can be installed on various Android versions without consuming much memory or crashing.

Keywords: deep learning, cloud, mobile application, ripeness, quality