A Multi-task Learning Approach for Weather Classification on Railway Transportation

Most of vision based urban transport dataset are designed to be executed in clear weather conditions. However, limited visibility in rain or cloudy strongly affects the accuracy of vision systems. To improve safety of railway transportation in actual weather situations, our newly constructed railway transportation dataset contains 4 situations from the real videos which has more adverse weather conditions. Taking into account railway transportation images are mainly single object with single background, which is limited to weather classification. We also collected a multi-class weather dataset to improve the generalization ability of the classification model. In order to capture a discriminate feature for each weather condition and avoid involving complicated pre-processing techniques. We provide a multi-task learning framework which formulate the classification problem as a multi-task regression problem by considering the classification on each weather class as a task. We capture the intrinsic relatedness among different tasks by a group Lasso regularization. With experiments on standard weather datasets and our own dataset, we demonstrate that the proposed framework achieves superior performance compared to the state-of-the-art methods.