Research and Analysis on the Application of Lightweight Algorithm Optimization in Complex Scenarios

With the rise of deep convolutional neural network and continuous development of artificial intelligence, remarkable achievements have been made in the field of target detection model. However, the traditional target detection algorithm based on deep learning usually needs huge computing resources and long-time training, which is difficult to deploy in edge computing devices, embedded devices, real-time monitoring and other environments. Therefore, the research and design of lightweight algorithm is one of the current research hotspots. This paper summarizes the operation effect of various lightweight algorithm optimization in complex scenes, aiming to further explore new ideas and new directions of lightweight network design. Firstly, the lightweight optimization strategies are divided into three categories: model pruning, knowledge distillation and low-rank Factorization. The performance, advantages and limitations of each kind of target detection task in complex environment are analyzed and introduced in detail. Finally, the research contents are summarized, and the challenges and potential directions of future work are discussed.