Deep Learning based Semantic Segmentation to Detect Ripened Strawberry Guava Fruits

Abstract

Strawberry Guava is a fruit that is high in nutrients. Manually harvesting this fruit is a time-consuming and labor-intensive task. The characteristics of ripened strawberry fruit need automated harvesting, as matured strawberries are unfit for consumption within two days. Deep learning-based approaches have arisen as answers to many issues in recent years. They offer a lot of hope in tricky sectors like agriculture, where they can manage distortion in data more successfully than the typical computer vision approaches. This paper describes a strawberry guava identification algorithm based on semantic segmentation. The modified UNet model was trained to segment ripened strawberry guava fruit with the help of human-annotated images appropriately. To analyze our experimental results on the segmentation of ripened strawberry guava the Dice score measure was used. The validation and test dataset dice scores were 91.04% and 89.72%. The proposed methodology demonstrated that matured strawberry guava could be accurately detected using the modified UNet semantic segmentation model with a few input images.