Weather conditions often disrupt the proper functioning of transportation systems. Multi-class outdoor weather classification is a difficult task to perform due to diversity and lack of distinct weather characteristics or features.
The aim of this project is to use transfer learning on the ResNet34 architecture and build an accurate classifier of still images of weather.
The data is generated as part of the research work by Ajayi, Gbeminiyi and the data is made accessible to the public for computer vision applications for detection and observation of weather conditions [1].
The dataset consists of a total of 1125 images which are maually labeled as cloudy, sunrise, rainy and sun shine. The distribution of the dataset is shown in the table below.
| Class | Cloudy | Sunshine | Rainy | Sunrise |
|---|---|---|---|---|
| Number | 300 | 235 | 215 | 357 |
A web app is hosted using the template from Render on the local machine and its preview is down below.
The image classifier built using CNN and ResNet34 architecture achieved an accuracy of 98.6%. This is achieved by transfer learning upon unfreezing the hidden layers and training using the above dataset.
[1] Ajayi, Gbeminiyi (2018), “Multi-class Weather Dataset for Image Classification”, Mendeley Data, V1, doi: 10.17632/4drtyfjtfy.1