基于迁移学习的植物病虫害细粒度分类

Fine-grained Classification of Plant Diseases and Pests Based on Transfer Learning

植物病虫害识别是农业生产中的一个重要问题。通过人工识别与传统计算视觉的方式在实际应用中有着一定的局限性,所以有必要研究基于卷积神经网络的植物病虫害分类算法。为了探讨卷积神经网络对于植物病虫害细粒度分类的应用,基于现有的部分植物病虫害数据图像数据,融合区分性区域注意力机制与通道间的注意力机制模型来评测此数据集,结果表明,融合注意力机制后相比于原始模型其分类性能有所提升,Top-1准确率提升了1.9%。为了进一步探究注意力机制对于卷积神经网络的影响,可视化了卷积神经网络关注的区域,注意力机制使得网络在细粒度分类任务中更关注于目标的区分性特征区域。

Identification of plant diseases and pests is an important issue in agricultural production.There are certain limitations in the practical application of manual recognition and traditional computational vision methods,so it is necessary to study plant disease and pest classification algorithms based on convolutional neural networks.In order to explore the application of convolutional neural networks in fine-grained classification of plant diseases and pests,based on existing partial plant disease and pest data image data,a discriminative region attention mechanism and inter channel attention mechanism model were fused to evaluate this dataset.The results showed that the classification performance was improved compared to the original model after fusing the attention mechanism,with a Top-1 accuracy increase of 1.9%.In order to further explore the impact of attention mechanisms on convolutional neural networks,the regions of attention of convolutional neural networks were visualized.The results showed that attention mechanisms make the network more focused on the discriminative feature regions of the target in fine-grained classification tasks.