Abstract：With the continuous development of deep learning in the field of object detection, the detection accuracy is constantly improved. However, the neural network algorithm model has high requirements on the computing resources of the hardware platform, so it is difficult to be applied in the embedded platform. In order to ensure that the neural network algorithm can meet the high accuracy and improve the efficiency of its operation, this paper carries out the research on the object detection algorithm based on neural network model compression technique. Firstly, K-means ++ clustering algorithm is used to cluster the prior box in the data set, in order to make algorithm have a good initial value in the initial stage of training. At the same time, aiming at the problem of computing speed, this paper optimizes the YOLOv3 which based on Darknet53 frame, and prunes the network model of YOLOv3.The experimental results show that the neural network model compression technique can improve the speed of the algorithm by two times with less loss of accuracy.
魏志飞, 宋泉宏, 李芳, 杨擎宇, 王爱华. 基于神经网络模型压缩技术的目标检测算法研究[J]. 空天防御, 2021, 4(4): 107-112.
WEI Zhifei, SONG Quanhong, LI Fang, YANG Qingyu, WANG Aihua. Research on Object Detection Algorithm Based on Neural Network Model Compression Technique. Air & Space Defense, 2021, 4(4): 107-112.
