摘 要：针对当前 Faster-RCNN 对小目标的检测精度较低，检测效果不佳等问题，提出一种改进算法。算法以 FasterRCNN 算法为基准算法，使用 Resnet101 残差网络代替传统 VGG16 网络，减少卷积操作，保留更多的中药目标有效特征信息，提取更多小目标的特征，并在算法中添加 CBAM 注意力机制处理策略来提高模型对中药检测的准确性，提高模型的检测精度。实验结果显示改进后的算法与传统算法相比在检测精度有 1.5% 以上的提升，证明了改进算法的鲁棒性。
关键词：Faster-RCNN；ResNet101；CBAM；中药检测
DOI:10.19850/j.cnki.2096-4706.2022.17.018
基金项目： 国家自然科学基金项目（61703005）；安徽省重点研发计划国际科技合作专项（202004b11020029）
中国分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）17-0071-04
Detection of Traditional Chinese Medicine Based on Improved Faster-RCNN
ZHAO Liuyang
(School of Computer Science and Engineering, Anhui University of Science and Technology, Huainan 232001, China)
Abstract: Aiming at the problems of low detection accuracy and poor detection effect of Faster- RCNN for small targets, an improved algorithm is proposed. The algorithm takes the Faster-RCNN algorithm as the benchmark algorithm, uses ResNet101 residual network to replace the traditional VGG16 network, reduces convolution operation, retains more effective feature information of traditional Chinese medicine targets, extracts more features of small targets, and adds CBAM attention mechanism processing strategy in the algorithm to improve the accuracy of traditional Chinese medicine detection and improve the detection accuracy of the model. Experimental results show that the detection accuracy of the improved algorithm is more than 1.5% compared with the traditional algorithm, which proves the robustness of the improved algorithm.
Keywords: Faster-RCNN; ResNet101; CBAM;CBAM; detection of traditional Chinese medicine
参考文献：
[1] TAO J H，GU Y H，SUN J Z，et al. Research on VGG16 Convolutional Neural Network Feature Classification Algorithm Based on Transfer Learning [C]//2021 2nd China International SAR Symposium （CISS）.Shanghai：IEEE，2021：1-3.
[2] ZHANG Z R.ResNet-Based Model for Autonomous Vehicles Trajectory Prediction [C]//2021 IEEE International Conference on Consumer Electronics and Computer Engineering （ICCECE）. Guangzhou：IEEE，2021：565-568.
[3] YU C Y，HE X，MA H T，et al. S-DenseNet：A DenseNet Compression Model Based on Convolution Grouping Strategy Using Skyline Method [J].IEEE Access，2019（99）：1.
[4] 陈艳菲，彭洪晟，王俊涛，等 . 基于 MobileNet 卷积神经网络的焊缝缺陷识别 [J]. 自动化与仪表，2022，37（1）：49-54.
[5] 李维刚，杨潮，蒋林，等 . 基于改进 YOLOv4 算法的室内场景目标检测 [J]. 激光与光电子学进展 .
[6] LU Y H，ZHANG L W，XIE W. YOLO-Compact： An Efficient YOLO Network for Single Category Real-Time Object Detection [C]//Proceedings of the 32th Chinese Control and Decision Conference（CCDC 2020）.Anhui：editorial board of Control and Decision，2020：1101-1106.
[7] 李琳，符明恒，张铁，等 . 基于改进 SSD 的工件定位算法[J]. 北京航空航天大学学报 .
[8] LIU R，YU Z H，MO D L，et al. An Improved FasterRCNN Algorithm for Object Detection in Remote Sensing Images [C]//2020 39th Chinese Control Conference （CCC）.Shenyang：TCCT，2020：7188-7192.
[9] 任丰仪，裴信彪，乔正，等 . 融合 CBAM 的 YOLOv4 轻量化检测方法 [J]. 小型微型计算机系统 .
[10] SUU K，SEIN M M. Event Analysis for Vehicle Classification using Fast RCNN [C]//2020 IEEE 9th Global Conference on Consumer Electronics（GCCE）.Kobe：IEEE，2020：2378-8143.
作者简介：赵留阳（1997—），男，汉族，河南固始人，硕士研究生在读，研究方向：目标检测。