<![CDATA[Purpose – This study aims to develop a semantic segmentation model for oyster mushrooms in cultivation media to support automated monitoring, growth analysis, and yield estimation in smart farming systems. Design/methods/approach – A Multi-Branch EfficientNet-U-Net hybrid architecture was proposed, using EfficientNet-B0 as the encoder and a multi-branch fusion strategy to integrate multi-scale features from three encoder levels. The dataset consisted of 150 manually annotated oyster mushroom images collected from two cultivation sites under varying illumination, mushroom cluster density, and background texture. The model was evaluated using Intersection over Union (IoU) and Dice Coefficient metrics on training, validation, and testing subsets. Findings – Experimental results show that the proposed model achieved high segmentation performance, with a median IoU of approximately 0.90 and a Dice coefficient of 0.93. Compared with the baseline U-Net, the proposed architecture produced cleaner segmentation boundaries and more consistent detection of mushroom regions under complex environmental conditions. Research implications/limitations – This study is limited by the relatively small dataset and evaluation on images from only two cultivation sites. Further studies should involve larger and more diverse datasets to assess robustness across broader cultivation environments. Originality/value – This study offers an effective semantic segmentation framework that combines EfficientNet-B0 encoding and multi-branch multi-scale feature fusion to improve oyster mushroom segmentation accuracy in real cultivation settings, with potential application in smart agriculture monitoring systems.]]>
Copyrights © 2026
