基于卷积神经网络的目标检测

Target detection based on convolutional neural network

为了完成视频图像中人体检测,将卷积神经网络和梯度算法结合。将YOLO网络模型结构以及算法函数进行了优化调整。为了保留视频时间信息,引入了卷积神经网络反馈时间维度。采用长短记忆卷积神经网络(LSTM)进行训练,通过将迭代过程的最优解保留,加快了梯度算法中的搜索速度,并且保证收敛于全局最优解。

To detect the human body in video images,we combine the convolutional neural network with gradient algorithm.The YOLO network model structure and its algorithm are optimized.In order to retain the time information of the video,the feedback time dimension of the convolutional neural network is introduced.We use both long and short memory convolutional neural network(LSTM)for training.By reserving the optimal solution in the iterative process,the search speed of the gradient algorithm is accelerated and global optimal solution is convergent.