Comparative Study of Traditional Convolutional Neural Network (CNN) And MobileNet Architecture for Weather Detection

This research is a comparative study of Traditional Convolutional Neural Network (CNN) andMobileNet architecture for effective weather condition detection. An open-source dataset of high-qualityweather images captured from different weather conditions was used for this study. The method adopted inthis study involves the training and validation of CNN and MobileNet models to classify weatherconditions such as cloudy, foggy, rainy, snowy, and sunny. The research objective is to compare theperformance of CNN and MobileNet models in accurately classifying weather conditions and identifyingthe strengths and limitations of each approach using metrics such as precision, recall and f1-score. Theresults indicate that while both CNN and MobileNet models achieved high overall accuracy in weatherclassification, there were differences in their performance across specific weather classes. CNN withaugmented images demonstrated superior speed and efficiency in inference, with an average accuracy of96% across the various weather classes, thereby making it well-suited for deployment on resourceconstrained devices. However, MobileNet models exhibited incredibly low accuracy, especially indetecting or properly classifying weather conditions when inputting new images. These findings provideinsights into the suitability of CNN and MobileNet architectures for weather classification tasks and offerguidance for future research and application development in this area.