Hao Zhiqiang, Wang Zhigang, Bai Dongxu, Tong Xiliang
Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan, China.
Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan University of Science and Technology, Wuhan, China.
Front Bioeng Biotechnol. 2022 Jul 1;10:945248. doi: 10.3389/fbioe.2022.945248. eCollection 2022.
Problems such as redundancy of detection model parameters make it difficult to apply to factory embedded device applications. This paper focuses on the analysis of different existing deep learning model compression algorithms and proposes a model pruning algorithm based on geometric median filtering for structured pruning and compression of defect segmentation detection networks on the basis of structured pruning. Through experimental comparisons and optimizations, the proposed optimization algorithm can greatly reduce the network parameters and computational effort to achieve effective pruning of the defect detection algorithm for steel plate surfaces.
检测模型参数冗余等问题使其难以应用于工厂嵌入式设备应用。本文重点分析了现有的不同深度学习模型压缩算法，并基于结构化剪枝提出了一种基于几何中值滤波的模型剪枝算法，用于对缺陷分割检测网络进行结构化剪枝和压缩。通过实验比较和优化，所提出的优化算法能够大幅减少网络参数和计算量，实现对钢板表面缺陷检测算法的有效剪枝。
