Abstract
Ultrasound signal processing plays an important role in medical image analysis. Embedded ultrasonography systems with low power consumption and high portability are suitable for disaster rescue, but due to the difficulty of ultrasonic signal recognition, operators need to have strong professional knowledge, and it is not easy to deploy ultrasonography systems in areas with relatively weak infrastructures. In recent years, with the continuous development in the field of deep learning and artificial intelligence, lightweight convolutional neural networks have brought new opportunities for ultrasound signal processing. This paper focuses on investigating lightweight convolutional neural networks applied to ultrasound signal classification. Combined with the characteristics of ultrasound signals, this paper provides a detailed review of lightweight algorithms from two perspectives: model compression and operational optimization. Among them, model compression deals with the overall framework to reduce network redundancy, and the latter aims at the lightweight design of the basic operational module "convolution" in the network. The experimental results of some classical models and algorithms on the ImageNet dataset are summarized. Through the comprehensive analysis, we present some problems and provide an outlook on the future development of lightweight techniques for ultrasound signal classification.