Deep learning model compression method and system for power equipment edge side recognition

Abstract

The invention provides a deep learning model compression method for power equipment edge side recognition, which realizes compression and acceleration of a related recognition model by directly modifying a trained deep learning model, and comprises the following steps: batch normalization layer fusion: fusing a batch normalization layer with a convolution layer before the batch normalization layer; for singular value decomposition of the full connection layer, adding an intermediate layer in front of the full connection layer based on a matrix singular value decomposition algorithm, and compressing parameters of the full connection layer; and quantization of model weight: performing quantization conversion on parameters by adopting a weight sharing method according to the redundancy of thedeep learning model. All the steps can be used independently and can also be matched with one another to work. The invention further provides a compression system. Compression and acceleration of thetrained deep learning model are realized, and the method and system have a wide application prospect in an environment of actively promoting ubiquitous electric power Internet of Things by a currentelectric power company.

Description

technical field [0001] The invention relates to the field of deep learning and the field of electric power ubiquitous Internet of Things, in particular to a deep learning model compression method and system for edge side recognition of electric power equipment for a trained power inspection or power image monitoring recognition model. Background technique [0002] The power system is an important pillar to ensure the stable development of the national economy. The overhead lines of the power system are large in scale, the surrounding environment is complex, and the climate changes frequently. In order to ensure the safe and stable operation of the power system and prevent accidents, fixed-point monitoring and regular inspections are required. [0003] Fixed-point monitoring adopts the method of installing a fixed camera device to monitor important areas in the power grid to prevent the invasion of foreign objects and idlers. The way of monitoring is usually to designate a s...

AI Technical Summary

Problems solved by technology

However, its huge amount of calculation and complex network structure make it i

Images