Commodity image retrieval method

• Main Authors: ZHANG SUOFEI, XING YANGYANG, LI CHUNXIANG, YANG MENGWEI, WU XIAOFU
• Format: Patent
• Language: Chinese; English
• Published: 05.05.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a commodity image retrieval method, which comprises the following steps of: establishing a commodity sample image recognition model, and training a deep convolutional neural network according to a commodity sample image with multi-semantic scale label information; after data enhancement operation is performed on the marked commodity sample image, inputting the marked commodity sample image into a convolutional layer of a deep convolutional neural network for feature extraction; specifying a deep convolutional neural network as a backbone network, copying the deep convolutional neural network and forming two branch networks; outputting the feature vector through a Softmax layer to obtain a commodity category sampling probability, and calculating a loss function value; and circulating the steps until a preset condition for training the deep convolutional neural network is achieved. Compared with the prior art, the performance of the commodity identification and retrieval system can be imp