FEM-Net:Feature-Enhanced Vision Mamba Network for camellia fruit segmentation

The natural growth conditions of camellia fruits are complex, and factors such as being obscured by branches and leaves, overlapping, and uneven light exposure can significantly affect the harvesting process, resulting in low efficiency. This paper focuses on the camellia fruit in natural settings and proposes a novel image segmentation algorithm for camellia fruits, called FEM-Net. First, we introduce the Mamba Attention Fusion block (MAF block) to enhance feature extraction, helping the model focus on the target areas and more accurately identify the morphological features of the camellia fruit. Additionally, we design the Global-Local Aggregation Module (GLA) to improve the model’s ability to perceive both local and global information, thus enhancing its capacity to capture the fine details of the camellia fruit. Experimental validation shows that FEM-Net achieves Precision, Recall, and F1 scores of 93.48%, 95.76%, and 94.52%, respectively, on the camellia fruit dataset. Compared to the baseline model, the Precision, Recall, and F1 scores improved by 3.74%, 3.60%, and 3.85%, respectively, outperforming other segmentation models.