Why Human Eyeball Grading Destroys Value
In traditional scrap yards from Delhi's Mayapuri to Mumbai's Dharavi, scrap quality evaluation is performed entirely by naked-eye guesswork. Aggregators deliberately exploit this visual ambiguity by claiming that clean aluminum alloy extrusions (such as 6063 architectural sections) are "contaminated with 30% zinc, iron screws, or severe oxidation."
This enables middlemen to deduct 25% to 45% tare penalties from the price paid to grassroots collectors. When the material is later sold to industrial smelters, the aggregator claims the exact opposite—declaring it to be "99% pure Grade A+ alloy" to capture an exorbitant arbitrage spread.
Agent 01 eliminates this human conflict entirely. By converting uncalibrated mobile smartphone photos and yard CCTV streams into an immutable, mathematically verifiable visual proof hash, quality becomes an objective physical property rather than a subjective negotiation point.
The 4-Stage Autonomous Vision Pipeline
Agent 01 does not rely on simple classification heuristics. It executes a multi-layer deep feature extraction pipeline designed specifically for chaotic, unsegmented scrap environments:
Stage 1: Semantic Foreground Isolation & Otsu Contour Thresholding
YOLOv8x-SegSeparates foreground scrap piles from concrete ground, mud, rubber tires, weighing scale pans, and human hands. Generates a polygon binary mask M(x,y) with sub-pixel boundary edge refinement.
Stage 2: Vision Transformer (ViT-B/16) Alloy Feature Extraction
768-Dim EmbeddingsPasses isolated scrap patches into a self-attention transformer network fine-tuned on 140,000 industrial scrap specimens (including extruded 6063, cast ADC12, berry copper wire, HDPE blow-molded bottles, and PCB circuit boards). Captures microscopic specular reflectivity and ductile tear textures.
Stage 3: CIELAB / HSV Color-Space Oxidation & Foreign Matter Profiling
ΔE* ColorimetryComputes Euclidean distance across CIE $L^*a^*b^*$ color space to identify iron rust ($Fe_2O_3$), copper patina ($Cu_2CO_3(OH)_2$), grease films, adhesive label residue, and dirt crusts. Calculates the exact foreign contamination surface ratio:
Stage 4: Deterministic ISO 9001:2015 Grade Mapping & Pinata IPFS Hashing
ERC-721 ProofMaps the composite purity score $\rho \in [0.00, 1.00]$ to standard industrial scrap tiers and generates an immutable IPFS SHA-256 content identifier pinned permanently to decentralized storage.
Standard Quality Classification Table
| Grade | Purity Threshold | Allowable Foreign Matter | MCX Price Factor | Permitted Industrial Application |
|---|---|---|---|---|
| Grade A+ | ≥ 98.0% | < 2.0% (Zero oil/grease) | 100% of Spot | Direct secondary induction smelting; extrusion billet forging |
| Grade A | 92.0% - 97.9% | 2.0% - 8.0% (Minor surface dust) | 94% of Spot | Automotive casting (ADC12), structural alloy ingots |
| Grade B | 85.0% - 91.9% | 8.0% - 15.0% (Paint/label residue) | 85% of Spot | Secondary de-oxidizer blocks, low-spec re-rolling mills |
| Reject | < 85.0% | > 15.0% (Excess moisture/soil) | Settlement Blocked | Requires mandatory pre-treatment, washing, or magnetic shredding |