Urbanization and increased vehicular movement have made parking management systems essential, with a key challenge being the optimal use of parking spaces, particularly during foggy seasons. Deep learning methods have proven effective for parking space detection, but hazy conditions exacerbate the difficulty. This research introduces a novel solution that combines Convolutional Neural Networks (CNN) with the Extreme Learning Machine (ELM) to improve accuracy while reducing training time. A specialized dataset, designed to classify parking spaces in foggy conditions, is used. In the proposed method, images are preprocessed using Light-DehazeNet (LD-Net) to enhance visibility. The CNN is then trained on the dehazed images to extract feature vectors, which are input into the ELM to classify parking spaces as occupied or vacant. Compared to standalone CNN, ELM, and other machine learning models, this approach shows superior accuracy and computational efficiency, making it highly suitable for real-time parking management in challenging conditions.