| Abstrakt: |
Strawberries are fragile, making them prone to various damages during the harvest and post-harvest process, leading to easy spoilage. To maintain quality, overripe or damaged strawberries must be sorted, but this is done manually by humans, making it subjective and inconsistent. By analyzing the color information from segmented strawberry flesh to determine its ripeness, it can be used as an objective measurement standard and can help with consistent selection. However, there are limitations in automating segmentation using a simple threshold-based method. Therefore, in this paper, strawberry flesh segmentation learning was conducted using a convolutional neural network to improve objectivity and automation of the selection process. To reduce information loss, pooling was not used, and instead employed a branch structure with convolutional layers of various filter sizes. Through this, it was confirmed that strawberries in various ripening states could be segmented more precisely and information loss could be reduced. [ABSTRACT FROM AUTHOR] |
