<![CDATA[The problem of waste management continues to increase along with population growth and lifestyle changes, highlighting the need for a fast and accurate waste classification system to support recycling processes. This study implements a transfer learning approach using seven Convolutional Neural Network (CNN) architectures: MobileNet, MobileNetV2, Xception, EfficientNetB0, VGG16, VGG19, and ResNet50 to classify waste into two categories: organic and recyclable. Each model is modified by adding a Global Average Pooling layer followed by a fully connected layer with 256 neurons before the output layer. The models are trained twice using 30 epochs, a batch size of 2, the Adam optimizer, and a learning rate of 0.0001. Experimental results show that ResNet50 achieves the best performance, with an accuracy of 89.84%, precision of 96.34%, recall of 82.82%, and an F1-score of 89.07%, followed by MobileNet with an accuracy of 89.25%. In contrast, Xception demonstrates the lowest performance, with an accuracy of 83.81%. Analysis of training and validation curves indicates that ResNet50 and MobileNet exhibit better stability and lower overfitting tendencies compared to other models.]]>
Copyrights © 2026
