Urban waste management faces challenges due to rising waste volumes and inefficient sorting methods. Traditional systems are time-consuming, error-prone, and less effective. This proposed work presents a smart waste management system combining convolutional neural networks (CNNs) and fuzzy logic to automate waste classification. The system processes grayscale images to categorize waste as hazardous, non-recyclable, organic, or recyclable. The CNN includes three convolutional layers with ReLU activation, max pooling, and fully connected layers with a softmax classifier. A Mamdani-type fuzzy inference system resolves classification ambiguities with “IF–THEN” rules. The CNNs and fuzzy logic integration improve accuracy, scalability, and cost-efficiency without requiring additional hardware like IoT sensors. In cases where the probabilities for recyclable and non-recyclable categories are similar, a mixed classification is assigned. The system’s reliability depends on predefined fuzzy rules, and its adaptability to new waste types may need further tuning. Trained on a dataset of 87,934 labeled images, the system achieved 95.73% accuracy, enhancing sorting speed and accuracy. This approach reduces the need for manual labor and improves recycling efficiency, promoting environmental conservation and sustainability. The primary source of errors was images that were blurred, where the model struggled to accurately classify the waste categories. The evaluation metrics include model accuracy and the execution time taken by each model. The system’s scalability offers potential for implementation in diverse urban settings.