Adaptive enhancement of spatial information in adverse weather

In the context of spatial information, particularly in video surveillance and intelligent transportation systems, the visibility of video images is severely impacted by adverse climates including rain, snow, and fog. Accurate and swift recognition of current weather conditions and adaptive clarification of surveillance videos are crucial to maintaining the integrity of spatial information. Addressing the limitations of traditional weather recognition methods and the scarcity of weather image datasets, a multicategory weather image block dataset was constructed. This research introduced a weather recognition algorithm that integrates image block processing with feature fusion. The algorithm uses traditional methods to extract shallow spatial features such as average gradient, contrast, saturation, and dark channel from weather images. It also employs transfer learning to fine-tune a pretrained VGG16 model, extracting deep spatial features from the model’s fully connected layers. The approach improves the SoftMax classifier’s recognition of fog, rain, snow, and clear weather photos by merging shallow and deep spatial information. This improvement is essential for the quality and reliability of spatial data in bad weather. The algorithm achieves 99.26% accuracy in weather detection; however, the best accuracy archive by state of art is 97.14%, confirming its usefulness as a module for adaptive video picture clarification in spatially informed systems.