Timely and accurate urban change detection is essential for supporting sustainable urban development, infrastructure management, and disaster response. Traditional 2D change detection approaches overlook vertical and structural alterations in dense urban settings. This study focuses on 3D multiple change detection in urban areas using high spatial resolution remote sensing imagery and digital surface models (DSMs) from two different time points, enabling the identification of both horizontal and vertical transformations. To address the complexity of urban changes, we developed a deep learning-based framework centered on Convolutional Neural Networks (CNNs) with various encoder architectures and customized loss functions. The input data consists of stacked multi-temporal optical images and corresponding DSMs, allowing the model to learn both spectral and elevation features. As part of a comparative analysis, we also implemented several traditional methods, including Principal Component Analysis (PCA), Change Vector Analysis with thresholding (CVA-thresholding), K-Means clustering, and a Random Forest classifier. An experimental evaluation was conducted on a high-resolution urban dataset, and performance was assessed using the F1-score, overall accuracy, and precision. The CNN-based models significantly outperformed the traditional methods, particularly in detecting complex and subtle structural changes that were otherwise missed. The best CNN model achieved an overall accuracy of 96.9%, an F1-score of 96.87%, and a recall of 95%. The integration of DSMs proved effective in capturing elevation-related changes, contributing to improved model sensitivity and classification performance. In conclusion, the proposed deep learning framework demonstrates strong potential for robust 3D urban change detection by effectively combining spectral and elevation data. While traditional methods offer useful baselines, the CNN-based approach provides superior accuracy and spatial detail, making it highly suitable for real-world urban monitoring applications.