Affected by the continuous advancement of science and technology, environmental changes, and the evolution of the times, the development of e-commerce has gradually become mainstream. The amount of commodities on the e-commerce platform is large. How to find the commodities the user is looking for is its most important function. The commodity search engine is the key to this function on the ecommerce platform. The traditional way which is the most common method of product retrieval is to use the keywords entered by the user to search for products in a large amount of data in text mode. However, to find the exact product in a large number of products through text retrieval rely on only text descriptions. There may exist semantic gaps among the input keywords, the user’s intention, and the description of the product itself. Using a content-based image retrieval system as the commodity searching engine, users may more easily find the product they want. We established a product search engine based on content-based image retrieval. After using the United Nations Standard Products and Services Code (UNSPSC) as the definition of product categories, the product images on momo shopping online are downloaded as the database for the learning of product classification and the target image database in retrieval. The learning for classification was based on the convolutional neural network to extract image features. Then, the extracted image features were used to match the similarities between the query product image and the target image database. The retrieval result using DenseNet had a mean average precision of 0.93. Using a deep convolutional neural network as the image retrieval base can achieve effective commodity search.