An outdoor weather image classification method based on a transfer learning technology

Abstract

The invention discloses an outdoor weather image classification method based on a transfer learning technology. The method comprises the following steps: firstly, establishing a deep network model, and then converting the trained deep network model in a checkpoint format into a deep network model in a pb format; and finally, judging the weather type of the image quickly and automatically only by inputting the image to be judged into the deep network model in the pb format. Classification accuracy is high, meanwhile, according to the outdoor weather image classification method, weather types can be classified; ready-to-classify fog, rain, sun and snow, meanwhile, subdivision can be carried out according to snowy weather; that is to say, subdivided into snows, medium snow and heavy snow, meanwhile, based on the transfer learning technology, the method has the advantages of being small in training data demand, short in model training time and high in optimization speed, the excellent deepnetwork ResNet is shown on an ImageNet data set in a transfer mode, and the high classification accuracy is obtained. The method is suitable for popularization and application in the technical fieldof image processing.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an outdoor weather image classification method based on transfer learning technology. Background technique [0002] Weather information is of great significance to our daily life, transportation and industrial production. Traditional weather information acquisition methods rely on various expensive weather data acquisition sensors, and also need to be judged by manual observation. In recent years, with the rapid development of image recognition technology, it has become a popular method to use ordinary cameras to obtain images for weather discrimination. [0003] The rapid development of technologies such as automatic driving, intelligent monitoring, and video image understanding has also driven the development of outdoor weather condition recognition technology, making it one of the current research hotspots. In the automatic driving system, the weather environment ar...

AI Technical Summary

Problems solved by technology

However, the deep learning model relies on large-scale labeled image data during training, which also leads to longer model training time and weaker generalization ability of the model.