Real-time detection algorithm for digital meters based on multi-scale feature fusion and GCS

Aiming at the problems of insufficient feature fusion, large number of network parameters and low target saliency in the current digital meter detection algorithm, a digital meter detection algorithm based on YOLOv5s is designed. First, a new feature fusion structure Bi-Directional Feature Pyramid Network Based on Multi-Scale Feature Fusion is designed to realize the full fusion between different scale feature maps and improve the detection accuracy; second, a new convolutional module Ghostconv Combined Channel Shuffle, is designed to realize the lightweight design of the network and meet the requirements of real-time substation detection tasks; finally, to improve the network’s ability to characterize the instrumented digits, the Convolutional Block Attention Module is introduced in the backbone network to further enhance the network performance. Experiments are carried out on the homemade dataset, and the experimental results show that the algorithm proposed in this paper improves the average accuracy by 2.84–98.58% compared with the original network; the amount of network parameters is reduced by 26.4%, and the detection speed is improved by 25 FPS, and the detection time for each image is only 0.012 s. Compared with other digital meter detection algorithms, the network performance and the number of parameters also have a great advantage.