Implementation of Mask Region-Based Convolutional Neural Network for Web-Based Waste Classification

Waste management is a critical issue in modern society, especially with the increasing challenges brought by population growth and urbanization. This research aims to develop a technology-based system that can automatically classify waste types, particularly distinguishing between organic and inorganic waste. The proposed model uses the Mask R-CNN algorithm with ResNet-50 as the backbone and is trained on a dataset of 4,570 images that underwent preprocessing through augmentation, resizing, and annotation using the Roboflow platform. The developed system allows users to upload waste images and returns outputs in the form of bounding boxes, classification labels, and segmentation masks for each detected object. Experimental results show that the best-performing model configuration in experiment 8 achieved excellent detection and classification performance. The evaluation using mean Average Precision (mAP) resulted in 96.67% for bounding boxes and 94.88% for segmentation masks. Moreover, the classification performance evaluated using the confusion matrix yielded an accuracy of 99.58%, precision of $99.60 \%$, recall of $99.60 \%$, and $\mathbf{F 1}$-score of $99.60 \%$, indicating the model’s outstanding ability to differentiate waste types. Despite the promising results, the model still faces limitations such as restricted background variety and object types within the dataset, which may affect its generalization in real-world scenarios.