基于改进DeepLab V3+模型的番茄叶片病害检测与识别研究

Research on Tomato Leaf Disease Detection and Recognition Based on Improved DeepLab V3+Model

番茄叶片病害的准确识别对于番茄病害防治至关重要。提出了一种基于改进DeepLab V3+语义分割模型的番茄叶片病害多类别分割模型,有效提升了番茄病害防治的精确性。研究中,为进一步提升特征的表达能力,在特征提取骨架的输出特征图之后引入自注意力模块,同时使用一种基于多层级通道注意力的特征融合策略,通过全局池化技术来捕捉通道间的相关性,改善了不同层级特征融合时的信息对齐问题。通过在番茄叶片病害数据集上的试验验证,改进后模型的平均像素准确率、平均交并比均有一定的提升,论证了改进的有效性。

Accurate identification of tomato leaf diseases is essential for tomato disease control.A multi-category segmentation model of tomato leaf diseases based on the improved DeepLab V3+model was proposed to improve the accuracy of tomato disease control.To improve the features expressive ability,a self-attention module was introduced after the output feature map of the backbone.A feature fusion strategy based on multi-level channel attention was also used,which captures the correlation between channels through global pooling and improves the information alignment problem during feature fusion at different levels.Through experimental verification on the tomato leaf disease dataset,the average pixel accuracy and average intersection and union ratio of the improved model have been improved to a certain extent,which demonstrates the effectiveness of the improvement.