IEEE Access · 2018 · 632 citations · 26 references
Convolutional Neural NetworkPrecision AgricultureEngineeringMachine LearningAgricultural EconomicsPlant PathologyDisease DetectionMaize Leaf ImagesPlant HealthMaize Leaf DiseaseAgricultural CyberneticsImage ClassificationImage AnalysisData SciencePattern RecognitionSustainable AgriculturePublic HealthMaize Leaf DiseasesMachine Learning ModelDeep LearningComputer VisionCrop Protection
Automatic identification of maize leaf diseases is highly desired in agricultural information systems. This study proposes improved GoogLeNet and Cifar10 deep‑learning models to enhance maize leaf disease identification accuracy while reducing network parameters. The authors refined the models by tuning parameters, modifying pooling, adding dropout and ReLU, and reducing classifiers, then trained and tested them on nine maize leaf image classes. The enhanced models achieve 98.9% (GoogLeNet) and 98.8% (Cifar10) top‑1 accuracy on eight disease classes, use far fewer parameters than VGG and AlexNet, and converge faster, improving training and recognition efficiency.
In the field of agricultural information, the automatic identification and diagnosis of maize leaf diseases is highly desired. To improve the identification accuracy of maize leaf diseases and reduce the number of network parameters, the improved GoogLeNet and Cifar10 models based on deep learning are proposed for leaf disease recognition in this paper. Two improved models that are used to train and test nine kinds of maize leaf images are obtained by adjusting the parameters, changing the pooling combinations, adding dropout operations and rectified linear unit functions, and reducing the number of classifiers. In addition, the number of parameters of the improved models is significantly smaller than that of the VGG and AlexNet structures. During the recognition of eight kinds of maize leaf diseases, the GoogLeNet model achieves a top - 1 average identification accuracy of 98.9%, and the Cifar10 model achieves an average accuracy of 98.8%. The improved methods are possibly improved the accuracy of maize leaf disease, and reduced the convergence iterations, which can effectively improve the model training and recognition efficiency.
26
Going deeper with convolutions
Christian Szegedy, Wei Liu, Yangqing Jia et al. · 2015 · 46.2K citations
Image Classification, Deep Neural Networks, Image Analysis +15
ImageNet Large Scale Visual Recognition Challenge
Olga Russakovsky, Jia Deng, Hao Su et al. · International Journal of Computer Vision · 2015 · 39.5K citations
Image Classification, Convolutional Neural Network, Machine Vision +7
Dropout: a simple way to prevent neural networks from overfitting
Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky et al. · 2014 · 34.2K citations