基于深度残差网络的茶园杂草分类及模型压缩方法研究

为提高茶园杂草分类深度模型的准确性,减少深度模型的冗余参数问题.以茶园常见的10类杂草图像为数据样本,分别基于深度学习的ResNet50、VGGNet和AlexNet网络结构构建杂草分类模型;在此基础上,进一步利用剪枝算法压缩深度模型ResNet50.通过实验对比3个模型测试集的平均准确率分别为0.86、0.72和0.63;此外,通过对比ResNet50的茶园杂草模型在训练集和测试集上压缩前后效果,显示结果基本一致.研究表明ResNet50在这3个模型中是最优分类模型,且压缩后的深度模型ResNet50提升了模型的性能.因此,该研究也为移动端设备的分类提供了理论基础.

Research on the traceability system of tea quality and safety based on blockchain

In order to improve the accuracy of the deep model for weed classification in tea gardens, and reduce the redundant parameter of this model. This paper uses 10 kinds of weed photographs taken in the garden to investigate the classification of weeds using different deep learning models (ResNet50, VGGNet, AlexNet), and on this basis, further applies pruning algorithms to compress the deep model ResNet50. Through experimentation, the average accuracy of these three models on the test set is 0.86, 0.72, 0.63; also, by comparing the effects of ResNet50 model before and after compression, the results were almost the same on the training set and the test set. Those shows the ResNet50 model is the optimal classification model among these three models, and the compressed deep model ResNet50 improves the performance of the model. Therefore, this research also provides a theoretical basis for the deep model for the classification of mobile devices.