Xception Network for Weather Image Recognition Based on Transfer Learning
Na Hyeon An · 2022 International Conference on Machine Learning and Intelligent Systems Engineering (MLISE) · 2022
Weather conditions are closely related to our lives. Traditional methods have some problems, such as small scale of weather image data set, difficult to optimize the model, and similar image factors of different weather types, which may lead to recognition errors. Therefore, this paper proposes a weather image recognition algorithm based on transfer learning and uses Xception algorithm to realize the network architecture. The Xception model was improved by adding a full connection layer, Dropout layer and Softmax layer. Combined with the functional requirements of weather image recognition, the identification of six different weather images was realized and compared with other models under the same parameter background. According to the results, the final accuracy of this method is up to 94.19%, which can effectively solve the problems of insufficient training samples and low accuracy in data sets.