Visual identification method for multiple types of articles based on neural network and metering equipment

• Main Authors: HUANG GUOLI, YANG CUIMEI, ZHANG ZHEN, LIN RONGTAO, LIN JUNRAN
• Format: Patent
• Language: Chinese; English
• Published: 07.01.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to a visual identification method for multiple types of articles based on a neural network. The method comprises the steps of obtaining N article image sets, and extracting category features corresponding to all article types; inputting the category features into a constructed first neural network for training to obtain a coarse recognition neural network; constructing N second neural networks, and inputting each article image set into each second neural network for training processing in pairs to obtain N fine recognition neural networks; obtaining a to-be-recognized image of a to-be-recognized article, performing coarse recognition on the to-be-recognized image through the coarse recognition neural networks, and determining a target article type of the to-be-recognized article; determining a fine recognition neural network according to the target article type; and performing fine recognition on the to-be-recognized image through the current fine recognition neural network, and determin