Nowadays, the number of goods in the market is increasing rapidly. In order to realize the efficient circulation of e-business products, it is necessary to apply the state-of-the-art image recognition technologies to improve the service quality. In this paper, on the basis of the deep learning towards the characteristics of product image, we formulate a novel deep architecture for searching products in e-business context. Specifically, we have designed a product image recognition platform, which consists of various actual cases. We have analyzed the application of this platform. Meanwhile, this work also discusses the theoretical basis of deep learning technology in time series forecasting, and sums up the commodity sales forecasting as multivariable time series. Using the historical sales data of an e-commerce online store, we also formulate the mechanism of constructing the LSTM network model under the well-known Tensorflow framework. We evaluate our method by comparing it with diverse AR models, it can be concluded that LSTM network model is simple and with convenient data input. Besides, it can be shown that good marketing decisions and reasonable inventory management is significant in products purchasing.

如今，市场上的商品数量正在迅速增加。为了实现电子商务产品的高效流通，需要应用最先进的图像识别技术来提高服务质量。在本文中，在对产品图像特征进行深度学习的基础上，我们制定了一种新颖的深度架构，用于在电子商务环境中搜索产品。具体来说，我们设计了一个产品图像识别平台，该平台由各种实际案例组成。我们分析了这个平台的应用。同时，本文还讨论了深度学习技术在时间序列预测中的理论基础，将商品销售预测总结为多变量时间序列。使用电子商务网上商店的历史销售数据，我们还制定了在著名的 Tensorflow 框架下构建 LSTM 网络模型的机制。我们通过与各种 AR 模型进行比较来评估我们的方法，可以得出结论，LSTM 网络模型简单且数据输入方便。此外，可以证明良好的营销决策和合理的库存管理在产品采购中具有重要意义。