基于Inception_ResnetV2网络模型的服饰属性识别

Clothing attribute identification based on Inception_ResnetV2 network model

为解决服饰属性识别困难的问题,文中在Inception_ResnetV2深度网络模型的基础上,先采用预训练模型的方法对数据进行预训练,在卷积层后面接入全局平均池化替代传统的全链接层再连接分类器,来解决训练模型造成的过拟合的问题。同时采用Adam优化器,降低模型收敛速度,最终实现不同种类服饰属性的识别。为验证模型的优越性,与Xception、InceptionV4等深度网络模型的训练结果相比,Inception_ResnetV2具有更高的识别率。从而证明基于Inception_ResnetV2的深度网络模型在服饰属性上具有更好的识别能力。

In order to solve the problem of clothing attribute recognition difficulty,based on the Inception_ResnetV2 deep network model,the pre-training model is used to pre-train the data,and then the global average pooling is used to replace the traditional full-link layer classifier after the convolution layer to solve the problem of over-fitting caused by the training model.At the same time,the Adam optimizer is used to reduce the convergence speed of the model.Finally,the identification of different kinds of clothing attributes is realized.In order to verify the superiority of the model,Inception_ResnetV2 has a higher recognition rate than the training results of deep network models such as Xception and InceptionV4,what proves that the deep network model based on Inception_ResnetV2 has better recognition ability in apparel attributes.