Multiclass Waste Image Classification Based on Convolutional Neural Network and Transfer Learning

Arif Rahman (Universitas Ahmad Dahlan)
Devaldi Caliesta Octadiani (Universitas Ahmad Dahlan)
Imam Riadi (Universitas Ahmad Dahlan)
Suprihatin (Universitas Ahmad Dahlan)

Journal of Innovation Information Technology and Application (JINITA)
Vol 8 No 1 (2026): JINITA, June 2026

The escalating waste generation presents significant challenges for waste management systems, driving the need for innovative and efficient waste separation technologies. Efficient waste management requires the accurate separation of organic and inorganic waste, which automation can significantly enhance. This paper uses Convolutional Neural Network (CNN) architectures, optimized through transfer learning, to classify waste images as part of an automatic waste separation system. The study evaluates three architectures (InceptionV3, ResNet152, and MobileNetV3) by fine-tuning and testing them on the Waste Images Dataset and Waste Classification Data datasets. The InceptionV3 model demonstrated superior performance on the Waste Images Dataset, achieving the highest accuracy across nine classes. In comparison, the MobileNetV3 model excelled on the Waste Classification Data with the best accuracy across the two classes. These results highlight the effectiveness of CNN in automating waste classification and underscore the potential of InceptionV3 for multiclass tasks and MobileNetV3 for binary classifications in diverse environmental contexts.