YOLO Computer Vision Developer for AI-Assisted Microscope Slide Screening (Asbestos Fibers)

I'm an environmental consultant building an AI-assisted screening tool for polarized light microscope (PLM) slides of bulk building-material samples. The tool triages slides — it scans, flags likely asbestos fibers, and presents them to a trained analyst for confirmation. It is a screening aid, not a certified analytical device. Hardware (I handle all physical setup): - Olympus BH-2 BHS trinocular polarizing microscope (I mount the camera myself) - AmScope MU503 5MP camera (USB 3.0) - DIY stepper-motor conversion of the manual XY stage and fine-focus knob (NEMA 17 motors, TMC2209 drivers, Arduino UNO) — design the motion-control layer around direct serial control of this DIY stage Scope: 1. Camera pipeline — live view and still capture from the AmScope camera in a desktop app (Windows). 2. Slide scanning — drive the motorized XY stage in a tile grid (via Micro-Manager or direct serial control), capturing overlapping fields across the slide. 3. Autofocus — motorized fine-focus control with software autofocus (focus-measure per field and/or focus-map interpolation across the slide), so every tile is captured sharp at 100x–400x. 4. Fiber detection + classification — train a YOLO (v8 or newer) model to detect fibers and classify: chrysotile, amosite, crocidolite, tremolite, actinolite, anthophyllite vs. non-asbestos fibers/debris. I will provide labeled slide images; advise on how many images per class you need and the labeling workflow. 5. Analyst review UI — tiled scan viewer with detection overlays; analyst confirms/rejects each detection; semi-automated point-count / visual area estimate of asbestos percentage per sample. 6. Report output — one-page PDF summary per sample (sample ID, fields scanned, confirmed fiber types, estimated percentage, analyst name/date). ### Phase 2 — PCM airborne fiber counting (NIOSH 7400) + mold spore detection 6. **Automated fiber counting** — detect and measure fibers on PCM filter slides per NIOSH 7400 rules (fibers longer than 5 µm, aspect ratio 3:1 or greater); count per Walton-Beckett graticule field, fibers/mm², and calculated airborne concentration (f/cc) from pump volume. Analyst reviews and adjusts counts. 7. **Mold spore detection** — detect and classify mold spores on spore-trap air samples (Air-O-Cell type cassettes): common genera such as Aspergillus/Penicillium-like, Cladosporium, Stachybotrys, Alternaria, and others vs. background debris/pollen. Spore counts per trace with analyst confirm/reject, same review UI as fiber screening. - Hardware note: PCM needs phase-contrast optics (phase objectives + phase condenser) — separate scope from the PLM (to be sourced); the camera and software pipeline carry over. - The PCM scope gets the same autonomous motorized stage treatment as the PLM scope: tile-grid scanning plus software autofocus, so both scopes scan slides hands-free. - Understood limitation: PCM cannot distinguish asbestos from non-asbestos fibers — this is a counting aid for air samples, not an identification tool. Milestones: 1. Camera + stage-control prototype: live view, autofocus, tile scan, image stitching (fixed-price or hourly — propose) 2. Labeled dataset review + YOLO training pipeline (advise on image counts first) 3. Trained PLM model integrated into the review UI with confirm/reject workflow 4. Point-count/area estimation + PDF report export; handover with setup docs 5. Phase 2: PCM fiber-counting module per NIOSH 7400 (separate estimate) Required skills: YOLOv8/v11 object detection, custom training on client datasets; Python, OpenCV; camera SDK integration (DirectShow / vendor SDK). Bonus: Micro-Manager, serial stage control, microscopy imaging. Engagement: Hourly contract. Please include in your proposal: (a) your estimated hours per milestone, (b) how many labeled images per fiber class you'd want, (c) any experience with microscope or lab imaging.