Urban Waste Profiling and Reporting System using Mask R-CNN Model
Year: 2026
Type: article
Abstract: Garbage piles up faster than any Indian municipal-ity can deal with it. The usual approach to understanding what people throw away still involves workers tearing open bags by hand, which is both slow and genuinely hazardous. Most attempts to automate the process have relied on convolutional neural networks that slap a single label onto an entire photograph. That is adequate for a picture of one bottle on a table. It is useless for a picture of actual roadside garbage where a squashed cola can sits on top of yesterday’s newspaper next to a torn polythene bag. What you really want is a system that can look at the heap and individually outline each piece of rubbish, separately. This paper describes EcoMask, a system we built around Mask R-CNN to do exactly that. We trained on the TACO litter dataset for 20,000 iterations, threw in aggressive augmentation (random flips, brightness jitter, ±15° rotations), and ended up with a bounding-box AP50 of 27.60% and a segmentation AP50 of 25.29%, with a mean inference time of 141.59 ms per frame on an NVIDIA T4 GPU. On the front end, a Flutter app lets citizens photograph waste, get disposal advice mapped to BBMP bin categories, and submit GPS-tagged reports. On the back end, a React dashboard gives municipal administrators a dual-mode heatmap with a one-kilometre radius scanner for planning targeted cleanups. The entire backend sits inside a Docker container on Hugging Face Spaces, with MongoDB Atlas handling persistence and Cloudinary hosting images so the database does not choke on base64 blobs. Keywords Instance Segmentation, Mask R-CNN, Solid Waste Management, Smart Cities, Deep Learning, Geospatial Analysis.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7213671762