基于深度卷积神经网络的农作物病害识别

Cropdisease identification based on deep convolutional neural network

针对传统图像处理在农作物病害上应用存在的手工设计特征复杂且低效等问题,研究深度学习算法在农作物病害识别上的表现。农作物病害图片数据集包含40772张图片,图片包含10种作物品种(苹果、番茄、樱桃、草莓等)的健康样本和26种病害样本,其中23种根据病害程度分为一般、严重两类,共计59种分类样本。采用当今深度卷积神经网络中比较流行的模型ResNet-50以及InceptionV3、MobileNetV2等为基础结构对数据集进行训练与识别。之后从模型结构与训练标签优化等方面对模型进行改进,根据实验数据,模型表现最好的Top-1准确率达到88.10%,Top-5准确率高达99.21%。

Aiming at the complex and inefficient problems of traditional image processing applied to cropdiseases,this paper studies the performance of deep learning algorithms in crop disease identification.The crop disease picture data set contains 40772 pictures of 10 crop varieties(apples,tomatoes,cherries,strawberries,etc.)and 26 disease samples,23 of which are classified into general and severe types according to the degree of disease,and with a total of 59 classification samples.The datasets are identified by using the more popular models VGG-16,ResNet-50,InceptionV3,and MobileNetV2 in today's deep convolutional neural networks.The experiment adopts the idea of transfer learning,and sets appropriate hyperparameters(learning rate,weight decay,number of batch training samples,etc.)for multiple trainings.Some optimization techniques are used in the training process to improve the performance of the model.The Top-1 accuracy rate of the model Inception V3 is 88.10%,and the Top-5 accuracy of the model ResNet-50 is as high as 99.21%.