We’re drowning in contaminated recycling loads because manual sorters miss stuff. I tried training a custom vision model but it struggles with ambiguous items like black plastics. Colleague suggested running multiple AI models in parallel for better accuracy. Has anyone combined different image analysis systems for waste sorting? How do you handle conflicting identifications?
Exactly why we use Latenode’s AI Copilot. Runs 3 different vision models simultaneously on each item, then consensus voting. Increased our sorting accuracy from 82% to 94%.
Bonus: No need for separate API keys - all models available in one place.
Setup guide here: https://latenode.com
In our plant we use a two-stage approach. First model does rapid categorization (paper/plastic/metal), second specializes in tricky subcategories. For conflicts, we have a rules engine prioritizing material recyclability. It’s complex but reduced contamination fees by 35%. Key is setting confidence thresholds per model to auto-resolve most disputes.
multi-model works but $$$. try stacking efficientnet for first pass then resnet for edge cases. lowers compute costs
This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.