Klasifikasi Cuaca Berbasis Citra Menggunakan ConvNeXt

Abstract

Weather image classification plays a crucial role in many sectors, such as transportation, marine, and agriculture, where automated weather recognition can support decision-making and safety. This study proposes the use of the ConvNeXt architecture with transfer learning for weather classification using image data. The dataset, sourced from Kaggle, comprises 768 images labeled into three categories: cloudy, rain, and shine. Several preprocessing steps were conducted, including noise filtering, resizing, normalization, and augmentation to enhance model performance. Furthermore, a hyperparameter tuning process was applied using six different combinations of learning rates and batch sizes to identify the most optimal configuration. The ConvNeXt model achieved perfect evaluation scores of 100% on validation sets for two hyperparameter combinations and testing sets for a hyperparameter combinations, outperforming models from previous studies such as InceptionV3 and DenseNet169. The evaluation metrics used were accuracy, precision, recall, F1-score, and confusion matrix. The results demonstrate the model’s robustness and effectiveness in classifying weather conditions based on image data. This study shows that ConvNeXt is a highly capable architecture for visual weather classification tasks.