<![CDATA[Waste management is an increasingly crucial environmental issue, particularly given the growing volume of waste and the lack of an effective sorting system. Reliance on manual sorting is considered inefficient, difficult to scale, and prone to errors, necessitating an automated approach based on visual intelligence. This study analyses the performance of two transfer learning-based Convolutional Neural Network (CNN) architectures, namely Xception and MobileNetV2, for image sorting of organic and inorganic waste. The Garbage Classification dataset, consisting of 15,515 images, was used, with preprocessing stages including normalisation, augmentation, handling class imbalance via class weights, and training using K-Fold Cross Validation and hyperparameter tuning. Validation results show that MobileNetV2 achieves the highest accuracy of 98.03%, but its performance decreases on the test data to 85.50%. In contrast, Xception demonstrates better generalisation with a test accuracy of 92.50%, an AUC of 0.918, and stable precision, recall, and F1-score metrics. A t-test also confirmed a statistically significant difference in the performance of the two models. Xception was deemed more feasible for implementation in an automated waste image sorting system under operational conditions. These results provide a basis for recommendations to developers and stakeholders to strengthen innovative waste management strategies and mitigate environmental impacts.]]>
Copyrights © 2026
