【摘要】 植物叶片对植物种类分辨与认知具有重大研究作用.提出了一种充分提取植物叶片特征信息的高阶信息增强模块，使用包含高阶信息增强模块的卷积神经网络模型对植物叶片图像进行多感受野特征提取.以复杂背景下的植物叶片图像为研究对象，从中国植物图像库中获取样本来源不同的植物叶片图像构成含有9种叶片的PLD＿amp数据集，采用添加高斯噪声、数据增广技术平滑和增扩数据集，增强数据的可操作性.与现有传统卷积网络相比，所提出的包含高阶信息增强模块的CNN模型最佳分类准确率可达88.7%,具有较高可行性与高分类准确率.更多还原

【Abstract】 Plant leaves play an essential role in the study of plant species discrimination and cognition. This paper proposes a high-level information enhancement module that fully extracts the feature information of plant leaves, and uses the convolutional neural network containing this module to extract the features of plant leaf images from multiple receptive fields. The experiment takes the plant leaf images in complex background as the research object, and plant leaf images from different sample sources are obtained from Plant Photo Bank of China(PPBC). These images constitute the PLD＿amp data set containing nine kinds of leaves. The techniques of adding Gaussian noise and data augmentation are used to smooth and expand the data set, thereby enhancing the data set’s operability. The CNN model’s best classification accuracy with a high-order information enhancement module proposed in this paper for plant leaf image classification in complex backgrounds reaches 88.7%. Compared with the existing traditional convolutional network, it has higher feasibility and classification accuracy, and it provides a new idea for plant leaf image recognition under complex background.更多还原