基于ResNet50的紫色土图像分类

Purple Soil Image Classification Based on ResNet50

针对紫色土图像数据集小、分类准确率不高的问题,提出一种基于ResNet50的小样本紫色土图像分类方法。首先,在ResNet50网络结构的最后一层卷积层后连接3层全连接层,采用SeLU激活函数,并加入Dropout层,构建紫色土图像分类模型;再引入迁移学习方法,用ImageNet数据集训练好的ResNet50网络参数初始化紫色土图像分类模型的卷积层参数,然后用紫色土图像数据集训练模型,微调模型参数,得到最终的紫色土图像分类模型。实验结果表明,基于ResNet50的紫色土图像分类方法在小样本紫色土图像数据集上能得到较好的准确率。

Aiming at the problem that the image dataset of purple soil is small and the classification accuracy is not high,proposes a small sample pur ple soil image classification method based on ResNet50.First,connects three fully connected layers after the last layer of the ResNet50 net work structure,uses the SeLU activation function,and joins the Dropout layer to construct a purple soil image classification model;Then,introduces the transfer learning method,initializes the convolution layer parameters of the purple soil image classification model with the ResNet50 network parameters trained by ImageNet dataset,trains the model with the purple soil image dataset,fine tunes the model param eters,and obtains the final purple soil image classification model.The experimental results show that the purple soil image classification method based on ResNet50 can get better accuracy on the small sample purple soil image dataset.