Research on the Application of Mask2Former for Segmenting Small Mushroom Caps of Seafood Mushrooms

In the task of monitoring mushroom growth characteristics, the mushroom cap is one of the most important features. To address the challenges faced by traditional instance segmentation algorithms in segmenting the caps of seafood mushrooms during the early stages of growth, such as unclear edges and low success rates. This paper proposes the application of the Mask2Former algorithm for instance segmentation of the caps of seafood mushrooms in their early differentiation stage. First, camera equipment was set up to capture images of the mushroom house for dataset creation. Then, the collected images were augmented through techniques such as flipping and adding noise. Finally, the processed images were input into the model, allowing Mask2Former to be compared with three other instance segmentation algorithms. Experimental results demonstrated that Mask2Former performed better in segmenting small mushroom caps during the early growth stage of seafood mushrooms, achieving an average precision (AP) of 96.731%, which effectively meets practical production needs.