Mamba-Based Super-Resolution and Segmentation Network for UAV-Captured Blueberry Farmland Imagery

• Published in: 2024 6th International Conference on Data-driven Optimization of Complex Systems (DOCS) pp. 644 - 649
• Main Authors: Zhao, Fan, Song, Jian, Wang, Jiaqi, Chen, Yijia, Xi, Dianhan, Shao, Xinlei, Liu, Y ongying, Ma, Bangzhang, He, Yinyin, Zhang, Guochen, Wu, Qingyang, Tu, Zhengyue, Wu, Mengya, Chen, Yulun, Guo, Jiaolin, Zhang, Chenyu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16-08-2024
• Subjects: Accuracy; Blueberry; Deep learning; Flowering plants; Image reconstruction; Magnification factor; Manuals; Optimization; Plantations; Semantic segmentation; Shape; Super-resolution reconstruction; Superresolution; UA V; Visualization
• DOI: 10.1109/DOCS63458.2024.10704386

Traditional methods of inspecting blueberry maturity primarily rely on manual observation, which is inefficient and prone to visual errors. The high similarity in shape and color between immature blueberries and blueberry flowers poses significant challenges for accurate visual assessment. This study proposes a novel method for evaluating the overall maturity of a blueberry plantation using super-resolution reconstruction (SRR) and semantic segmentation. We designed and evaluated various SRR models, including SRCNN, EDSR, RCAN, Real-ESRGAN, SwinIR, HAT, and MambaIR, to enhance the detailed representation of blueberries at different maturity stages in low-resolution images. These enhanced images are then processed by the proposed semantic segmentation network to accurately segment blueberry flowers, leaves, and blueberries at different maturity stages. Experimental results demonstrate that MambaIR achieves 82.26% SSIM in the SRR task, while the image segmentation model developed based on the Mamba algorithm attains an average intersection over union (mIoU) of 80.48. In summary, the proposed method shows significant accuracy and efficiency advantages in assessing blueberry maturity.