فحص الجودة بالرؤية الحاسوبية

المتطلبات تصف قدرات يحتاجها النمط في بيئتك، مو الموردين اللي لازم تشتريهم. أي نظام يملأ متطلباً يحقّقه — وهذا اللي يخلّي الكتالوج قابلاً للنقل عبر الذيل الطويل من أدوات الشركات الصغيرة والمتوسطة.

1. image_capture_source

   مطلوبقراءةبث

   Where images of items come from. Physical setup matters: lighting, angle, consistency.

   شكل البيانات

   Image or short video clip with item identifier, capture timestamp, line/station identifier.
   يُملأ عادةً بواسطة

   • camera installed over a production line
   • imaging station integrated with the conveyor system
   • handheld scanner used by inspectors
   • phone-based capture at pack-out stations
2. defect_taxonomy_and_examples

   مطلوبقراءةمجموعة

   What 'bad' looks like, specifically. The pattern learns from examples; without enough labeled examples per defect class, it can't be trusted.

   شكل البيانات

   Per-defect-class: name, description, decision rule, set of labeled positive and negative example images.
   يُملأ عادةً بواسطة

   • labeled dataset built during the engagement
   • ongoing labeling workflow where QC inspectors add to the dataset
   • structured defect catalog the operations team maintains
3. item_record_lookup

   مطلوبقراءةطلب

   What item is being inspected and what its specifications are. Different items have different inspection criteria.

   شكل البيانات

   Per-item SKU with inspection specification, packaging requirements, customer-specific variations.
   يُملأ عادةً بواسطة

   • ERP item master with QC specifications
   • product catalog with inspection criteria
   • customer-specific shipment requirements
4. inspection_decision_destination

   مطلوبكتابةحدث

   Where pass/fail decisions go. Pass-through for accepted items, divert for rejected items.

   شكل البيانات

   Decision per item: accept/reject/uncertain, defect class if rejected, confidence score, image stored for review.
   يُملأ عادةً بواسطة

   • control signal to the production line diverter
   • scan log in the warehouse management system
   • result feed into the operations dashboard
5. human_qc_review_station

   مطلوبقراءة + كتابةطلب

   Where uncertain or borderline cases get human verification. Critical for the cases the pattern doesn't handle confidently.

   شكل البيانات

   Image with proposed classification, confidence score, comparison to similar cases. Reviewer marks final decision.
   يُملأ عادةً بواسطة

   • dedicated review station at the QC desk
   • tablet at the inspection point with the review UI
   • remote review queue for off-line review
6. audit_image_archive

   مطلوبكتابةمجموعة

   Stored images of inspected items for traceability and quality investigations. Particularly important in regulated industries.

   شكل البيانات

   Image with item identifier, inspection result, timestamp, retention metadata.
   يُملأ عادةً بواسطة

   • object store with retention policies
   • QC archive in the manufacturing operations system
   • compliance archive for regulated products
7. model_retraining_loop

   مطلوبقراءة + كتابةدفعة

   How human verdicts feed back to improve the pattern. Without this loop, the model can't adapt to new defect types or environmental changes.

   شكل البيانات

   Reviewed cases with original prediction, human verdict, image, reason for override.
   يُملأ عادةً بواسطة

   • weekly batch of reviewed cases added to the training set
   • continuous labeling pipeline where each correction feeds back
   • quarterly model retraining process