摘要
针对SSD目标检测模型参数量大、运行速率低的问题,在SSD模型的基础上提出一种新的煤矸快速识别模型DSR-SSD.应用深度可分离卷积代替主干特征提取网络中的普通卷积,减少了模型的计算量;将RFB模块融入到SSD模型中,提高了模型的特征提取能力.经验证,DSR-SSD模型的识别速率为113.99帧/s、精确率为95.17%.将DSR-SSD与SSD,Faster-RCNN,YOLOv3三种模型对比,发现DSR-SSD模型与SSD模型相比,精确率提高了2.29%,识别速率提高了60.89%;同时,DSR-SSD模型的精确率比Faster-RCNN模型高2.86%,比YOLOv3模型高2.71%,识别速率分别是Faster-RCNN模型和YOLOv3模型的14.90倍和3.65倍,证明了DSR-SSD模型性能优越.
A new model DSR-SSD for coal gangue fast identification is proposed based on the SSD model to address the issues of large parameter quantities and low operating speed in the SSD model.The application of deep separable convolutions in the backbone feature extraction network reduces the computational complexity,and integrating the RFB module into the SSD model improves the model’s feature extraction ability.After verification,the recognition rate of the DSR-SSD model is 113.99 frames/s,and the accuracy rate is 95.17%.Comparing DSR-SSD with SSD,Faster-RCNN,and YOLOv3 models,it is found that the DSR-SSD model improves the accuracy by 2.29%and the recognition rate by 60.89%compared to the SSD model,and the accuracy of the DSR-SSD model is 2.86%higher than the Faster-RCNN and 2.71%higher than the YOLOv3,with recognition rates 14.90 and 3.65 times higher than the Faster-RCNN and YOLOv3.
作者
李娟莉
魏代良
李博
文小
LI Juan-li;WEI Dai-liang;LI Bo;WEN Xiao(College of Mechanical and Vehicle Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Shanxi Key Laboratory of Fully-Mechanized Coal Mining Equipment,Taiyuan University of Technology,Taiyuan 030024,China)
出处
《东北大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2023年第10期1474-1480,共7页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(51875386,51804207).
关键词
煤矸分选
深度学习
目标检测
SSD模型
轻量化
objective
interval particle
interval coal gangue separation
deep learning
target detection
SSD model
light-weighting
