Shiitake Mushroom Semantic Segmentation Method Based on Search Focus Network

Abstract

The substantially similar texture features of sticks and shiitake mushrooms in the mushroom-growing environment make precisely labeled samples more expensive and semantic segmentation of shiitake mushrooms more challenging. In this paper, a search focus network(SFNet) for semantic segmentation of shiitake mushrooms was proposed, which utilized the group-reversal attention module(GRAM) to strengthen semantic information understanding and trained via transfer learning and data augmentation strategies. The experimental results on the self-built shiitake mushroom sticks dataset revealed that structural measure $S_{\alpha}$, weighted F-measure $F_{\beta}^{\omega}$, adaptive E-measure $E_{\phi}^{ad}$, and absolute mean error $M$ of SFNet were 0.9161, 0.9113, 0.9808, and 0.0049, respectively, with practical and steady performance. With only a few training samples, the proposed approach can accomplish the semantic segmentation task of shiitake mushrooms.