Road condition detection based on adaptive random forest migration learning
In order to improve the accuracy and efficiency under small datasets,a road image classification method based on improved convolutional neural network model VGG16 and adaptive random forest transfer learning is proposed.Transfer learning was carried out on the improved VGG16 model,and the convolutional layer,pooling layer,and fully connected layer of the VGG16 network trained on a large dataset were trans-ferred.The random forest classification algorithm was used to replace the softmax layer of the VGG16 network for re-learning and training,overcoming the disadvantage of softmax's emphasis on the independence between features.In addition,the quantum revolving door update strategy of the quantum wolf pack algorithm is im-proved and has been used for the random forest hyperparameter optimization to ensure that the random forest is transferred with the best parameters to train and further improve the generalization ability of the model.The ex-perimental results show that in the image classification experiments which is self-built and provided by Kaggle website,the image recognition accuracy is 98.08%,and the classification speed has also been significantly im-proved.
road condition detectiontransfer learningquantum wolf pack algorithmadaptive random forestconvolutional neural network
NETL
NSTL
万方数据
