基于SSD卷积网络的视频目标检测研究

针对传统卷积神经网络对远距离视频目标识别效果差的原因,本文提出一种改进的基于SSD卷积网络的视频目标检测模型.首先,对数据集进行剪裁,旋转等预处理,提高网络检测泛化能力,其次,采用coco数据集Mobilenet_SSD预训练模型,由于其具有轻量级网络模型特点,减少计算开销,减少内存占用量.然后,再结合voc2012数据集进行二次训练微调处理,加快训练收敛速度,使用自定义数据集能有效检测特定场景目标,能够有效识别远距离场景下视频目标物体.实验结果表明,改进的网络检测模型适用于远距离目标检测,减少计算量,降低硬件内存资源消耗,提高网络模型性能和检测精确度,具有较好的鲁棒性.

Research on Video Target Detection Based onSSD Convolution Network

Aiming at the poor effect of traditional convolutional neural network on remote target recognition,an improved video target detection model based on SSD convolution network is proposed in this paper.First,data sets are tailored,rotated and preprocessed to improve the generalization ability of network detection.Secondly,the coco data set Mobilenet_SSD pre training model is adopted,which has the characteristics of lightweight network model,reducing computation cost and reducing memory usage.Then,combined with the voc2012 data set,the study does two training fine-tuning processes to accelerate the convergence speed of training,and uses custom data sets to detect specific scene targets effectively,and effectively identify objects in long distance scenes.The paper proposes a SSD convolution based target detection network model,which is suitable for long distance target detection and improves the performance of network models through retraining fine tuning scale,and has strong applicability.