Research On Weather Recognition Method Based On Multi-convolution Ensemble Learning

With the rapid development of various industries,timely and accurate identification and prediction of weather conditions has become a very urgent task.Especially in the field of transportation,the uncertainty of weather changes poses a great risk to road safety.In the context of the rapid development of artificial intelligence,the image-based weather phenomenon recognition classification has solved the defect that the original weather recognition method has a wider recognition range and cannot be specific to a certain location.At the same time,it is based on pictures taken by cameras everywhere on the road.Weather identification ensures the real-time nature of the judgment results,which makes people more efficient in handling various emergencies.The most common method of weather classification by pictures is to first extract the characteristics of various weathers in the pictures,and then select the classifier to classify the weather.The similarity of the characteristics of weather pictures is high,and the selection of multiple classifiers and the classification process are extremely complicated,which is not conducive to a large number of future applications and lacks universality.In view of the above weather classification problem,this paper proposes a weather image recognition method based on multi-convolution integrated learning.Convolutional neural network(CNN)is a simple and fast image classification method that plays an important role in the advancement of artificial intelligence.Compared with other machine learning methods,the convolutional neural network automatically learns deeper semantic features of the image through the convolutional layer and the pooling layer,which makes the machine more in line with the characteristics of human image recognition in the image classification process,and greatly shortens Classification time.Evidential Reasoning(ER)rules are a powerful method of uncertainty reasoning and information fusion,which can play an important role in the decision-making stage of artificial intelligence.In this paper,based on the structural framework of convolutional neural network,aiming at the defects of the uncertainty of deep learning in the decision-making stage,the method of integrated learning is used,and four convolutional neural network models are selected for data sets containing four types of weather: sunny,cloudy,rain,and snow Perform parallel recognition,input the obtained four results into ER-Rule for fusion,and obtain the final recognition accuracy.The whole process includes three stages of perception(image acquisition),cognition(CNN classification),and decision-making(ER-Rule fusion),which is closer to the human recognition method.Through experimental verification,the accuracy rate obtained by this method is compared with the four accuracy rates obtained by the recognition method using only convolutional neural networks.The accuracy rate result is improved by up to 4%,which proves the availability of the method.