Published Paper Details:

Turbid media, such as haze, smoke, fog, rain, or snow, typically impair image data collected by outside visual sensors. As a result, weather conditions would typically impair or impede the proper operation of vision-assisted transportation systems, or ADAS (advanced driver assistance systems), as well as a variety of other outside surveillance-based systems. To address these issues, the removal of weather effects (or deweathering) from photographs has become more significant and has gained a lot of attention. As a result, it is critical to include a preprocessing stage that automatically determines the weather state for an input image, after which the appropriate deweathering processes (e.g., removal of haze, rain, or snow) are correctly triggered. This study proposes a deep learning-based weather image recognition framework that takes into account the three most typical weather variables in outdoor scenes: hazy, wet, and snowy. Our approach automatically classifies an input image into one of the three categories or none of them (e.g., sunny or others). Extensive experiments on well-known CNN models such as VGG-19 and Inception V3 conclude that VGG-19 provides more than 96% accuracy when compared to Inception V3, and thus Vgg-19 is declared as the best model for performing the application on jehanbhathena/weather-dataset to evaluate the proposed method and the feasibility has been verified.