摘要
针对城市停车难,车位检测环境复杂等情况,研究了一种基于Inception-V4算法的车位状态检测方法。在Inception-V4网络结构基础上使用Leaky_ReLU代替ReLU作为激活函数,解决ReLU激活函数引起的神经元失活问题;在网络分类层前添加FReLU激活函数层和多个全连接层,使其获得有更丰富语义信息的特征向量,防止了网络过拟合问题,提高车位状态检测模型的整体性能。基于PKLot停车场数据集的实验结果表明,该方法对车位状态检测准确率较原模型有较大程度的提升。
Due to the difficulty of parking in cities and the complex detection environment of parking spaces,a parking space state detection method based on Inception-V4 algorithm is studied in this paper.On the basis of the Inception-V4 network structure,Leaky_ReLU is used instead of ReLU as the activation function to solve the problem of neuron inactivation caused by the ReLU activation function;the FReLU activation function layer and multiple fully connected layers are added before the network classification layer to obtain feature vectors with richer semantic information,which prevents network over-fitting problems and improves the overall performance of the parking space state detection model.The experimental results on PKLot parking lot data set show that the accuracy of the parking space state detection of this method is greatly improved compared with the original model.
作者
王栋
蔡斌斌
宰昶丰
Wang Dong;Cai Binbin;Zai Changfeng(College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao,Shandong 266590,China)
出处
《计算机时代》
2022年第3期5-10,共6页
Computer Era
