Research map / Large language models / AI and society
XCotton: Advancing AI-Enabled Hardware/Software Integrated System for Foreign Fiber Cleaning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents XCotton, an AI-enabled hardware/software system with a line-scan camera and a few-shot CLIP-based network (FSFFD) for detecting and removing foreign fibers in cotton, deployed in about 100 factories.
Problem
Foreign fibers (hair, hemp rope, plastic film, polypropylene rope) contaminate cotton during picking and processing and are hard to remove, lowering cotton quality, especially in developing countries.
Approach
Combines a line-scan camera with white and red/green laser illumination and a few-shot CLIP-based detector that learns object-agnostic text prompts with global and local (multi-layer) context optimization, integrated into a deployed lint cleaning machine.
Key results
- The proposed FSFFD method reaches 92.9% accuracy and 86.2% IoU, versus 83.2% / 63.8% for CLIP, 77.2% / 61.9% for UNet, and 70.3% / 56.3% for SVM.
- Ablation: adding local-only optimization gives 88.7% accuracy and 75.3% IoU; both losses combined give 92.9% and 86.2%.
- Deployed machine reports 1000 kg/h cleaning capacity (43% improvement) and 14 kWh energy use (63% less), with over 100 orders by Aug 2024.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use XCotton instead of CLIP | Foreign fiber detection on the collected dataset · Collected dataset | Acc: 83.2 -> 92.9 | Helped |
| Use XCotton instead of CLIP | Foreign fiber detection on the collected dataset · Collected dataset | IoU: 63.8 -> 86.2 | Helped |
| Add global context optimization (ablation) | XCotton · Collected dataset | Acc: 83.2 -> 86.1 | Helped |
| Add global context optimization (ablation) | XCotton · Collected dataset | IoU: 63.8 -> 69.2 | Helped |
| Add local context optimization (ablation) | XCotton · Collected dataset | Acc: 83.2 -> 88.7 | Helped |
| Add local context optimization (ablation) | XCotton · Collected dataset | IoU: 63.8 -> 75.3 | Helped |
| Add global context optimization while keeping local context optimization (ablation) | XCotton · Collected dataset | Acc: 88.7 -> 92.9 | Helped |
| Add global context optimization while keeping local context optimization (ablation) | XCotton · Collected dataset | IoU: 75.3 -> 86.2 | Helped |
| Add local context optimization while keeping global context optimization (ablation) | XCotton · Collected dataset | Acc: 86.1 -> 92.9 | Helped |
| Add local context optimization while keeping global context optimization (ablation) | XCotton · Collected dataset | IoU: 69.2 -> 86.2 | Helped |
| Deploy XCotton instead of a traditional machine | Lint cleaning machine · Cotton processing | Processing Capacity: ≤700 kg/h -> ≥1000 kg/h | Helped |
| Deploy XCotton instead of a traditional machine | Lint cleaning machine · Cotton processing | Energy Consumption per 100kg: ≥4.1 kWh -> ≤1.5 kWh | Helped |
Limitations
- Evaluation uses a private in-house dataset; only 100 images were released.
- Business and deployment claims are self-reported by the manufacturer collaborator, with no independent verification.
Experimental setup
Datasets: in-house line-scan cotton images collected in production (over 10,000, about 3,000 with foreign fibers)
Benchmarks: in-house foreign fiber dataset (100 released images subset)
Metrics: accuracy IoU processing capacity (kg/h) energy consumption (kWh) recognition rate
Baselines: SVM UNet CLIP
Builds on: CLIP CoOp-style context optimization focal loss dice loss
Closest papers by meaning
- MicroFiberDetect: An Application for the Detection of Microfibres in Wastewater Sludge Based on CNNs (AAAI 2025 · similarity 0.74)
- Towards Efficient and Intelligent Laser Weeding: Method and Dataset for Weed Stem Detection (AAAI 2025 · similarity 0.73)
- Infusing Synthetic Data with Real-World Patterns for Zero-Shot Material State Segmentation (NeurIPS 2024 · similarity 0.73)
- Digital Twin-Driven Teat Localization and Shape Identification for Dairy Cow (Student Abstract) (AAAI 2024 · similarity 0.72)
- AgMMU: A Comprehensive Agricultural Multimodal Understanding Benchmark (NeurIPS 2025 · similarity 0.71)
- STING-BEE: Towards Vision-Language Model for Real-World X-ray Baggage Security Inspection (CVPR 2025 · similarity 0.71)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.