文献类型： 会议论文

第一作者： Zili Chen

作者： Zili Chen ¹ ; Yilong Peng ¹ ; Yaokai Yang ² ; Jiadong Jiao ¹ ; Laigang Wang ³ ; Aiguo Wang ⁴ ; Jianjun Liu ⁵ ; Wei Lin ⁶ ; Yan Guo ⁷ ;

作者机构： 1.College of Computer and Information Engineering, Henan Normal University, Xinxiang, China

2.School of Tourism, Henan Normal University, Xinxiang, China

3.Institute of Agricultural Information Technology, Henan Academy of Agricultural Sciences, Zhengzhou, China

4.Zhengzhou Tobacco Research Institute of CNTC, Zhengzhou, China

5.Henan Provincial Tobacco Company of CNTC, Zhengzhou, China

6.College of Computer and Information Engineering, Henan Normal University, Xinxiang, China|Henan Provincial Key Laboratory of Educational Artificial Intelligence and Personalized Learning, Xinxiang, China

7.Institute of Agricultural Information Technology, Henan Academy of Agricultural Sciences, Zhengzhou, China|Key Laboratory of Huang-Huai-Hai Smart Agricultural Technology, Ministry of Agriculture and Rural Affairs, Zhengzhou, China

关键词： Location awareness;Accuracy;Shape;Convolution;Signal processing algorithms;Logic gates;Feature extraction;Robustness;Lesions;Diseases

会议名称： International Conference on Intelligent Computing and Signal Processing

主办单位：

页码： 1785-1791

摘要： For tobacco leaf disease spots with varying scales and shapes causing difficulties in segmentation and low accuracy, a novel MD_Unet algorithm is proposed for precise segmentation of tobacco leaf disease spots, aiding in scientific prevention and control of tobacco diseases. Innovative design includes Multi-scale Convolution Module (MC) and Dense Residual Dilated Convolution Module (MVCR) to enhance feature extraction of multiscale disease spots and perception of global contextual information. Utilizing Residual Unit Fusion Framework (RUFF) in the upsampling section effectively compensates for spatial details loss and strengthens network representation capability. Additionally, integration of Channel Spatial Attention Mechanism (CBAM) and Attention Gate (AG) further enhances the network's focusing ability on key information and fine-grained feature extraction capability. The MD_Unet algorithm achieves significant improvements in spot CPA, Recall, IoU, F1 metrics with 92.75%, 90.94%,84.93%,91.81% respectively, and an overall Dice score of 94.67%. Compared to mainstream methods like Unet, PSP, DeepLab v3+, FCN, SegNet, UNET++, DoubleU-Net, it shows remarkable enhancements, providing theoretical basis and technical support for precise segmentation of tobacco leaf disease spots and other plant diseases.

分类号： tp3