Tobacco leaf disease hole detection method based on spectral–spatial fusion multimodal transformer

作者：Jie Zhang, Zhanhao Zhang, Dayong Xu, Jiakang Li, Hongfu Zhu, Duanfei Wang, Ruidong Li, Jinsong Du, Fengxian Wang · 发表于：Industrial Crops and Products · 年份：2026 · DOI：10.1016/j.indcrop.2026.123538 · 研究领域：Smart Agriculture and AI、Water Quality Monitoring and Analysis、Plant Surface Properties and Treatments

• Developed a CNN-Transformer dual-branch network for multi-modal tobacco defect detection. • Designed an adaptive fusion mechanism to enhance cross-modal feature consistency. • Built a dynamic detection framework to improve small lesion localization. • Introduced a multi-scale loss function to optimize learning for small targets.