We address the vision-based autonomous landing problem in complex urban environments using deep neural networks for semantic segmentation and risk assessment for Uncrewed Aerial Vehicles (UAVs). We propose employing the SegFormer for the semantic segmentation of the complex and unstructured urban environments. The assessment is done when the video feed from an RGB camera on the UAV is segmented into typical urban classes and then mapped to a level of risk, considering in general, potential damage to property, the drone itself or endanger people. This approach yields valuable information that can be leveraged when UAV missions need to land in urban spaces, especially when in emergency resulting from system failures or human errors. The proposed strategy is validated through several case studies, demonstrating the huge potential of semantic segmentation-based strategies to determine the safest landing areas for landing, which will help unleash the full potential of UAVs on civil applications within urban areas.