A Hybrid CNN Approach in Waste Classification

Recycling plays a vital role in environmental sustainability and the conservation of natural resources. However, increasing population, industrialization, and waste have increased the need for automation in waste sorting processes. This study investigated the performance of deep learning-based classification models on waste classification using two different datasets (Garbage Classification and Garbage-Dataset). Experiments were conducted on four Convolutional Neural Networks (DenseNet121, InceptionV3, MobileNetV2, and VGG16), utilizing data augmentation and transfer learning techniques. A hybrid model was created by aggregating the features of these four models. Performance evaluation was performed with the 5-fold cross-validation method. In both datasets analyzed according to the experimental results, the hybrid model yielded the highest performance metrics, including Accuracy, Precision, Recall, F1-score, and ROC-AUC. A test accuracy rate of 85.88% was obtained in the Garbage Classification dataset and 91.26% in the Garbage-Dataset. The study highlights the critical impact of dataset size and model architecture choices on classification performance, providing an essential foundation for developing automation solutions in waste management.