基于改进YOLO v3算法的零件目标快速检测方法

Fast Detection Method of Part Target Based on Improved YOLO v3 Algorithm

针对通用算法在小规模数据集时零件目标检测效率不高的问题,提出一种基于YOLO v3算法的改进网络.改进YOLO v3算法针对数据集的特点,在充分保留深度卷积网络特征提取能力的基础上,减少网络层数、模型参数量和检测特征图,能够有效提高算法检测速度.对比试验证明,改进YOLO v3算法与原YOLO v3算法相比,两者的检测效果没有明显差距,但改进YOLO v3算法的检测效率得到了明显提升.

Rapid detection of part targets is of great significance in production activities.To address the problem of low detection efficiency of the general algorithm in small-scale data sets,an improved network based on YOLO v3 algorithm is proposed.According to the characteristics of the data set,the improved algorithm fully retains the ability of deep convolutional network feature extraction,and reduces the number of network layers,model parameters and detection feature maps,which can improve the detection speed of the algorithm.The comparison experiments show that compared with the original YOLO v3 algorithm,the improved algorithm has no obvious difference in detection effect,but the detection efficiency of the improved algorithm has been significantly improved.