Convolutional neural network model compression method and system, and computer readable medium

• Main Authors: LIN HENGJIE, ZHOU LINGYA, LIU MIAO
• Format: Patent
• Language: Chinese; English
• Published: 04.11.2022
• 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 relates to a compression method and system of a convolutional neural network model and a computer readable medium. The convolutional neural network model is used for processing an image, and the compression method comprises the following steps: constructing the convolutional neural network model; performing forward reasoning on the image by using a convolutional neural network model to obtain a feature map label; performing pruning processing on the convolutional neural network model to obtain a pruned convolutional neural network model; and retraining the pruned convolutional neural network model according to the feature map label. According to the DBSCAN algorithm, pruning is carried out on the convolution kernel, model retraining is carried out in combination with the feature map label and the Wing Loss function, the pruning technology enables the model to obtain relatively large sparseness on the premise that the model precision is guaranteed, the calculation speed of the model is well impro