基于GoogLeNet Inception V3的迁移学习研究

Research on Transfer Learning Based on GoogLeNet Inception V3

随着人工智能的再度崛起,使用深度学习模型进行图像分类的方法得到了广泛关注。针对典型深度卷积神经网络模型是在大型数据库和大算力的基础上进行训练得到的,但普通机器学习工作者很难拿到如此规模的数据集和算力现象,本文在GoogLeNet Inception V3深度学习模型的基础上,对GoogLeNet的特征提取模块进行迁移学习来训练特定的模型进行图像分类。实验结果表明,在硬件和数据集相对不足的条件下,采用迁移学习的策略可以高效地实现目标检测。

With the re-emergence of artificial intelligence,the method of image recognition using deep learning models has received extensive attention.The typical deep convolutional neural network model is based on large-scale database and large computing power,but it is difficult for ordinary machine learners to get such database and computing power.Based on the GoogLeNet Inception V3 deep learning model,the GoogLeNet’s feature extraction module is introduced to perform transfer learning to train specific models for image classification.The experimental results show that the transfer learning strategy can achieve target detection efficiently under the condition that the hardware and data sets are relatively insufficient.