光学 精密工程, 2020, 28 (1): 251, 网络出版: 2020-03-25
改进YOLOv2卷积神经网络的多类型合作目标检测
Multi-type cooperative targets detection using improved YOLOv2 convolutional neural network
合作目标 目标检测 数据增强 改进YOLOv2 卷积神经网络 cooperation target object detection data augmentation improved YOLOv2 convolutional neural network
摘要
针对大型构件三维精密测量中构件结构复杂、测量环境变化等导致的合作目标检测精度低的问题, 提出一种改进YOLOv2卷积神经网络的多类型合作目标检测方法。首先, 利用WGAN-GP生成对抗网络扩增合作目标图像样本数量; 其次, 采用卷积层密集连接代替YOLOv2基础网络的逐层连接增强图像特征信息流, 引入空间金字塔池化汇聚图像局部区域特征, 构建改进YOLOv2卷积神经网络的多类型合作目标检测方法; 最后, 采用增强的目标图像样本数据集训练改进YOLOv2卷积神经网络的多类型合作目标检测模型, 实现多类型合作目标检测。实验结果表明: 采用多类型合作目标图像数据集测试, 多类型合作目标检测精度达到90.48%, 目标检测速度为58.7 frame/s。该方法具有较高的检测精度和速度, 鲁棒性好, 满足大型构件三维精密测量中多类型合作目标检测的要求。
Abstract
In the three-dimensional (3D) precision measurement of large component, the detection accuracy of cooperative targets is low due to complex structure of large components and various measurement environment. To solve this problem, a multi-type cooperative target detection method using improved YOLOv2 convolutional neural network was proposed. Firstly, the data augmentation method combined with WGAN-GP was employed to amplify the number of cooperative target images. Secondly, the convolutional layer dense connection was used instead of the YOLOv2 basic network layer-by-layer connection to enhance image feature information flow, and the spatial pyramid pooled was introduced to convergence image local area feature. Base on those two parts, the multi-type cooperative targets detection method with improved YOLOv2 convolutional neural network was constructed. Finally, the multi-type cooperative targets detection model with improved YOLOv2 convolutional neural network was trained by the augmentation dataset for detecting the multi-type cooperative targets. The experimental results of multi-type cooperative target detection indicate that, detection precision of the proposed method is up to 90.48%, and detection speed is 58.7 frame per second by using image dataset of multi-type cooperative targets to test. This method has higher precision, rapid speed and strong robustness, which can satisfy the multi-type cooperation targets′ detection requirements for 3D precision measurement of the large component.
王建林, 付雪松, 黄展超, 郭永奇, 王汝童, 赵利强. 改进YOLOv2卷积神经网络的多类型合作目标检测[J]. 光学 精密工程, 2020, 28(1): 251. WANG Jian-lin, FU Xue-song, HUANG Zhan-chao, GUO Yong-qi, WANG Ru-tong, ZHAO Li-qiang. Multi-type cooperative targets detection using improved YOLOv2 convolutional neural network[J]. Optics and Precision Engineering, 2020, 28(1): 251.