Classification Method of Significant Rice Pests Based on Deep Learning

Rice pests are one of the main factors affecting rice yield. The accurate identification facilitates timely preventive measures to avoid economic losses. Some existing open source datasets related pest mostly include only a small number samples, or suffer from inter-class and intra-class variance data imbalance challenges, which limit application deep learning techniques in field identification. In this paper, based on IP102 dataset, we first reorganized large-scale dataset for by Web crawler technique manual screening. This was given name IP_RicePests. Specifically, includes 8248 images belonging 14 categories. IP_RicePests then expanded 14,000 via ARGAN augmentation address difficulties obtaining large samples pests. Finally, parameters trained public image ImageNet using VGGNet, ResNet MobileNet networks were used as initial values target training network achieve classification experimental results show that all three combined with transfer have good recognition accuracy, among highest accuracy can be obtained fine-tuning VGG16 network. addition, following demonstrates high improvements models, obtains augmented dataset. It is demonstrated CNN employ overcome sample sizes improve efficiency study provides foundational technical support