作者：陈文韵1, 王学影1, 胡晓峰1, 郭斌2
作者单位：1. 中国计量大学计量测试工程学院，浙江 杭州 310018;
2. 杭州沃镭智能科技股份有限公司，浙江 杭州 310018
关键词：深度学习;efficientdet;卷积神经网络;目标检测
摘要：
针对传统的图像处理方法对于机械零件检测存在的检测时间长、准确率低等难点，提出一种基于efficientdet的汽车ecu分类检测方法，将经过预处理和数据增强的ecu外壳图片样本输入神经网络训练，利用一种改进的新型的加权双向特征提取网络bifpn和一种复合尺度扩张方法进行特征提取并匹配特征图，提高检测的准确率，利用预训练模型进行迁移学习缩减训练时长，实现ecu外壳的自动检测。将检测结果与 faster r-cnn、mask r-cnn、efficientdet-d0模型检测结果相比较，实验结果表明，基于efficientdet的机械零件检测算法的识别率高于对比的其他网络模型，map达92.4%，在实际应用中更能够精确地检测ecu零件，满足实验与生产线检测需求。
efficientdet based automotive ecu classification and testing method
chen wenyun1, wang xueying1, hu xiaofeng1, guo bin2
1. college of metrology and measurement engineering, china jiliang university, hangzhou 310018, china;
2. hangzhou wolei intelligent technology co., ltd., hangzhou 310018, china
abstract: in order to solve the problem of the traditional image processing method, it is difficult to detect machine parts with long detection time and low precision, and efficientdet-based ecu classification method was proposed. the pre-processed and data enhanced ecu shell samples were fed into the neural network training.an improved weighted bidirectional feature extraction network bifpn and a compound scale expansion method were used to extract features and match the feature map to improve the precision of detection. the pre-training model was used for transfer learning to reduce the training time, and the automatic detection of ecu shell was realized.the results were compared with the faster r-cnn, mask r-cnn and efficientdet-d0 models. the efficientdet-based algorithm achieved a better recognition rate than its peer networks, with map reaching 92.4%.in practical applications, it can more accurately detect ecu parts to meet the requirements of test and production line testing.
keywords: deep learning;efficientdet;convolutional neural network;target detection
2023, 49(1):98-104 收稿日期: 2021-07-07;收到修改稿日期: 2021-09-27
基金项目: 国家自然科学基金（52075511）
作者简介: 陈文韵(1997-),女,浙江温州市人,硕士研究生,专业方向为机器视觉、深度学习
参考文献
[1] 刘桂雄, 何彬媛, 吴俊芳, 等. 面向视觉检测的深度学习图像分类网络及在零部件质量检测中应用[j]. 中国测试, 2019, 45(7): 1-10
[2] 王昕, 赵飞, 蒋佐富, 等. 迁移学习和卷积神经网络电力设备图像识别方法[j]. 中国测试, 2020, 46(5): 108-113
[3] 安丽娜. 基于深度卷积神经网络的图像分类识别方法[j]. 信息通信, 2019(12): 72-73
[4] he k, zhang x, ren s, et al. deep residual learning for image recognition[c]. ieee conference on computer vision & pattern recognition. ieee computer society, 2016: 770-778.
[5] huang g, liu z, laurens v, et al. densely connected convolutional networks[c]. ieee computer society, 2017: 4700-4708.
[6] ren s, he k, girshick r, et al. faster r-cnn: towards real-time object detectionwith region proposal networks.[j]. ieee transactions on pattern analysis and machine intelligence, 2017, 39(6): 1137-1149
[7] tan m, le q v. efficientnet: rethinking model scaling for convolutional neural networks[c]. international conference onmachine learning. long beach, usa: icml, 2019: 6105-6114.
[8] tan m, pang r, le q v. efficientdet: scalable and efficient object detection[c]. proceedings of the ieee/cvf conference on computer vision and pattern recognition. 2020: 10781-10790.
[9] redmon j, divvala s, girshick r, et al. you only lookonce: unified, real-time object detection[c]. proceedings of the ieee conference on computer vision and pattern recognition, 2016: 779-788.
[10] 乔婷婷, 李鲁群. 结合数据增广和迁移学习的高分辨率遥感影像场景分类[j]. 测绘通报, 2020(2): 37-42
[11] 高竟博, 李晔, 杜闯. 基于深度学习的小龙虾分级算法[j]. 现代计算机, 2020(26): 40-46
[12] girshick r. fast r-cnn[c]. proceedings of the ieee international conference on computer vision, 2015: 1440-1448.
[13] lin t y, dollár p, girshick r, et al. feature pyramid networks for object detection[c]. proceedings of the ieee conference on computer vision and pattern recognition. 2017: 2117-2125.
[14] liu s, qi l, qin h, et al. path aggregation network for instance segmentation[c]. proceedings of the ieee conference on computer vision and pattern recognition, 2018: 8759-876.
[15] 包壮壮, 赵学军. 基于efficientdet的无预训练sar图船舶检测器[j]. 北京航空航天大学学报, 2021, 47(8): 1664-1672.
[16] 刘思洋, 黄坚, 刘桂雄, 等. alexnet两光照下多类别法定货币识别技术[j]. 中国测试, 2019, 45(9): 118-122
