基于卷积神经网络的农作物叶片病害诊断

Diagnosis of crop foliar diseases based on convolutional neural network

农作物病害诊断对于及时发现并采取防控措施具有重要意义。本研究针对苹果、玉米、番茄、葡萄等典型农作物的常见叶片病害问题,使用了两种目前使用最广泛的卷积神经网络——VGG16及Resnet50,对典型农作物叶片病害进行识别。使用AI Challenger比赛的农作物叶片病害数据集图像,并对这些图像进行预处理,构建了47285张图片的数据集。分析两种卷积神经网络的性能,实验结果表明:VGG16及Resnet50分别达到了82.57%和86.34%的准确率,且Resnet50收敛速度更快,更适合农作物叶片病害的诊断识别。

The diagnosis of crop diseases is very important for timely detection and control.Aiming at the common leaf diseases of apple,corn,tomato,grape and other typical crops,two convolutional neural networks,VGG16 and Resnet50,are used to identify the leaf diseases of typical crops.Using the images of crop leaf disease data set of AI Challenger competition,and pre-processing these images,47285 data sets were constructed.The performance of the two convolutional neural networks is analyzed.The experimental results show that the accuracy of VGG16 and Resnet50 is 82.57%and 86.34%respectively,and Resnet50 converges faster,which is more suitable for the diagnosis and recognition of crop leaf diseases.