基于迁移学习的番茄病虫害检测

Detection of tomato pests and diseases based on transfer learning

为对番茄病虫害叶片特征进行提取,减少番茄病虫害叶片图像模型复杂度,缓和卷积神经网络模型检测番茄病虫害叶片图像时出现的过拟合现象,提出一种利用迁移学习技术实现卷积神经网络的分类模型,利用训练成熟的卷积网络的多层结构将底层特征逐步提升为抽象的高层特征,使其具有良好的特征学习能力。实现番茄病虫害叶片图片的数据增强,对数据增强后的叶片图片进行特征提取,利用支持向量机对图片进行分类。实验结果表明,该方法在番茄病虫害检测中具有较高的准确性和鲁棒性。

To extract the leaf table features of tomato pests and diseases, to reduce the complexity of leaf table picture model of tomato pests and diseases, and to mitigate the occurrence of over-fitting phenomena when using the convolution neural network model detects the leaf table pictures of tomato pests and diseases, a classification model of convolutional neural network was proposed using the transfer learning technique, and the multi-layer structure of well-trained convolutional network was used to elevate the underlying features into abstract high-level features, so that it has good characteristic learning ability. The images of tomato pests and diseases were enhanced, feature extraction was carried out, and the images were classified using support vector machine. Experimental results show that the proposed method has high accuracy and robustness in tomato pest detection.