Image gradient-based convolution method, directional convolution-based neural network and classification method

• Main Authors: LIU MAOMEI, ZHONG SHENG, LUO HANGZAI, PENG JINYE, TANG LEI, KUANG NAILIANG
• Format: Patent
• Language: Chinese; English
• Published: 16.07.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention discloses an image gradient-based convolution method, a directional convolution-based neural network and a classification method, and belongs to the field of convolutional neural networks. According to the method, the image gradient information-based direction convolution is performed on the image to extract the image features. According to the neural network based on directional convolution, explicit priori knowledge-image gradient information is embedded into a deep learning model, the scale of a network parameter space is effectively reduced, and the problem of local extremum is reduced. According to the method, the prior knowledge is added on the basis of the shallow network to improve the precision, and the image classification recognition accuracy similar to that of a deep model is achieved by using a shallow model with few parameters. The directional convolution introduced miniaturized convolutional neural network model achieves classification precision similar to that of an existing dept