| 摘 要: 作为全球重要粮食作物,水稻的生长易受病害影响,精准识别水稻叶片病害对保障水稻的产量和质量至关重要。传统人工识别方法主观性强、效率低,难以满足农业智能化需求。针对雾霾等复杂环境下水稻叶片病害检测精度下降的问题,提出改进的MixDehazeNet-CNN架构,融合深度可分离卷积与条件参数化卷积(CondConv),构建自适应多尺度特征提取模块。实验结果表明,该方法在11类水稻病害识别中准确率达96.83%,较传统模型显著提升,具备良好的鲁棒性与应用潜力。 |
| 关键词: 水稻病害识别 深度学习 MixDehazeNet 雾霾环境 条件参数化卷积 |
|
中图分类号: S511;S435;TP391
文献标识码: A
|
|
| Rice Leaf Disease Detection Method Based on Improved Mix Dehaze Net Architecture with Conditional Parameterized Convolution |
|
SHI Rui, WU Lili
|
(College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
2013441958@ qq.com; wull@ gsau.edu.cn
|
| Abstract: As one of the world’s major staple crops, rice is highly susceptible to diseases that threaten yield and quality. Accurate rice leaf disease identification is therefore essential for ensuring stable production. Traditional manual diagnosis is subjective and inefficient, limiting its application in intelligent agriculture. To address the decline in rice leaf detection accuracy under complex environments such as haze, this study proposes an improved MixDehazeNe-t CNN architecture that integrates depthwise separable convolution and Conditionally parameterized Convolution (CondConv) to construct an adaptive mult-i scale feature extraction module. Experimental results show that the proposed method achieves an accuracy of 96. 83% in identifying 11 types of rice leaf diseases, significantly outperforming conventional models and demonstrating strong robustness and practical potential. |
| Keywords: rice disease detection deep learning MixDehazeNet conditional hazy environment parameterized convolution |
