A Step towards Efficient Waste Segmentation for Automated Waste Recycling in Cluttered Background

ICLR 2026 Conference Submission18517 Authors

Keywords: Waste Recycling, Multi-scale Feature Aggregation, Spectral Domain, Difference of Gaussian, Feature Enhancement, Semantic Segmentation

Abstract:

Rapid expansion of urban areas and population growth is causing an immense increase in waste production, which demands the need for efficient and automated waste management. In this scenario, automated waste recycling that utilizes deep learning methods to separate the recyclable waste objects may emerge as a savior to humanity. Recent deep learning approaches for automated waste recycling provide promising waste segmentation performance in cluttered scenarios. However, these methods rely on large backbone networks that are inefficient for automated waste recycling systems. To this end, we propose an efficient waste segmentation network, where the spatial context module enhances localized structural dependencies in the spatial domain, and the spectral context module subsequently captures global contextual relationships in the frequency domain. This cascaded design allows the network to progressively leverage both local and global representations across complementary domains to highlight the semantic information necessary for effective segmentation of waste objects in cluttered scenes.
Furthermore, our auxiliary feature enhancement focuses on structural information to enhance the target objects' boundaries and blob amplification for better segmentation.
Extensive experimentation on two challenging waste segmentation datasets including ZeroWaste-f and SpectralWaste reveals the merits of the proposed method.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 18517