Loading
AI visual recognition + laser inspection + barcode scanning — three independent verification channels fused in real time for ≥99.99% accuracy at 0.7 seconds per pack
In hospital pharmacies, every dispensed medication must be verified before it reaches the patient. A single error — the wrong drug, the wrong dosage, or the wrong patient — can have serious clinical consequences. According to published research on medication errors, dispensing inaccuracies occur at rates between 0.1% and 3.0% in manual processes, depending on workload, staff fatigue, and drug similarity.
Traditional manual verification requires a pharmacist to visually inspect each drug pack, checking the drug name, specification, quantity, and patient label against a prescription. This process typically takes 3 to 5 seconds per pack — manageable for small volumes, but increasingly strained as prescription volumes grow. During peak hours, a single pharmacist in a busy hospital may verify hundreds of packs per shift, making sustained attention difficult.
Several factors compound the challenge:
Drug verification in hospital pharmacies has evolved through three distinct stages, each adding capability but also revealing limitations that the next generation needed to address.
A pharmacist visually compares each drug pack against the prescription order. This approach relies entirely on human attention and experience. While flexible and requiring no equipment investment, it is inherently limited by fatigue, workload volume, and the similarity of many drug packages. Error rates increase significantly during high-volume periods or extended shifts. Manual checking processes 3 to 5 seconds per pack and provides no automated audit trail.
Barcode scanning introduced a first layer of automation by reading drug traceability codes and matching them against a database. This significantly reduced transcription errors and created digital records. However, barcode-only systems have a critical blind spot: they verify the code but not the physical drug. If the wrong drug is placed in a correctly labeled package, or if a barcode is damaged or misprinted, a barcode scanner alone will not catch the discrepancy. The system trusts the label, not the contents.
The current generation combines multiple independent verification channels — AI visual recognition, physical measurement, and code reading — into a single fused decision. Rather than relying on any one data source, these systems cross-validate drug identity through fundamentally different measurement principles. If one channel produces an ambiguous result, the others provide corroboration or contradiction. JKBEAT's tri-modal fusion technology represents this approach, integrating AI vision, laser inspection, and barcode scanning into one verification cycle.
JKBEAT's tri-modal fusion technology (AI visual recognition + laser inspection + barcode scanning) processes three independent data streams simultaneously, fusing them into a single verification decision within 0.7 seconds. Each modality examines the drug pack from a different physical principle, making the system resilient to conditions that would defeat any single-channel approach.
JKBEAT's fully self-developed AI vision algorithms — not a third-party module — analyze high-resolution images of each drug pack to identify four categories of visual features: printed text (drug name, specification, manufacturer), graphic elements (logos, regulatory marks), color patterns, and geometric shape. The system uses deep learning models trained on a proprietary dataset of drug packaging images, enabling it to distinguish between drugs with highly similar appearances.
Three specialized algorithms work in concert: one for surface defect detection, one for color verification, and one for size measurement. These correspond to three of JKBEAT's registered software copyrights, developed between January and July 2022. Together, they create a visual "fingerprint" of each drug pack that the system compares against its reference database.
While the AI visual system examines what the drug pack looks like, the laser inspection system measures what it physically is. A precision laser sensor scans each drug pack to determine its thickness profile, detecting subtle differences that visual inspection alone may miss. For blister packs, the laser can identify whether individual blisters contain tablets or capsules and whether any are missing. For bottled medications, it verifies seal integrity.
Laser inspection serves as a physical corroboration layer. If a drug pack has the correct label but contains the wrong number of units, or if a blister pack has a missing tablet, the laser system detects the anomaly regardless of what the visual channel reports. This independence from the visual channel is precisely what makes the tri-modal approach robust: each modality can catch errors that the others cannot.
The third modality reads the drug traceability code — either a linear barcode or a two-dimensional QR code — affixed to or printed on each drug pack. This code links the physical item to its digital record in the national drug traceability database. The scanner decodes the identifier and validates it against the expected prescription order, confirming that the correct drug specification has been selected for the correct patient.
With China's 2026 mandatory drug traceability policy requiring that every code be scanned and uploaded — "scan every code, upload every scan" — the barcode channel serves a dual purpose: it contributes to verification accuracy and simultaneously fulfills regulatory compliance requirements. The scan result is automatically recorded in the system's audit log, creating the traceable record that inspectors and hospital administrators require.
The strength of tri-modal fusion lies not in any single channel but in the decision logic that combines all three. When a drug pack enters the verification chamber, all three systems begin their analysis simultaneously:
These three scores are fused by the system's decision engine. A pack passes verification only when all three modalities independently confirm a match. If any one modality flags a discrepancy, the system rejects the pack, logs the reason (which channel failed and why), and alerts the operator with a detailed image and data report. This "all must agree" logic is what enables the system to achieve ≥99.99% recognition accuracy with zero missed detections — a single-channel failure cannot slip through because the other channels serve as independent checks.
The entire process, from pack entry to final decision, completes in 0.7 seconds per pack — 4 to 7 times faster than manual verification at 3 to 5 seconds per pack.
All performance figures below are based on JKBEAT's internal testing and field deployment data from hospital pharmacy environments.
Recognition Accuracy
With zero missed detections — every pack is verified
Per Pack Verification Speed
4–7 times faster than manual 3–5 seconds
Drug Specifications Supported
Continuously expanding reference database
Packaging Types Supported
Blister packs, bottles, boxes, ampoules
The combination of ≥99.99% accuracy with zero missed detections reflects the tri-modal architecture's design principle: no single verification channel is the sole arbiter of pass or fail. The zero missed detection rate means the system does not allow a drug pack to pass through unexamined — every item that enters the verification chamber is processed by all three modalities. Accuracy refers to the correctness of the verification result (correct match or correct rejection), while zero missed detections refers to the system's guarantee that no pack bypasses inspection entirely.
JKBEAT is certified under GB/T19001-2016/ISO9001:2015 Quality Management System (Certificate No. CZ2511009QR1, issued by Shanghai Chenzun Certification Co., Ltd.). The certification scope covers the assembly and production of drug packaging machines, verification machines, and unpacking machines. First certified on November 19, 2022; the current certificate was issued on November 17, 2025, and remains valid until November 16, 2028.
JKBEAT holds a granted invention patent: "A Medical Verification Machine Facilitating Drug Transport" (Patent No. ZL202410362593.X, Publication No. CN117963604B), filed March 28, 2024, and granted July 5, 2024. Inventor: Wan Jun. This patent protects the core mechanical and operational design of the drug verification machine, underscoring the company's commitment to self-developed technology.
JKBEAT has registered five computer software copyrights covering the core algorithms and control systems: AI Visual Recognition Object Surface Defect Algorithm System V1.0, AI Visual Recognition Object Color Algorithm Program V1.0, AI Visual Recognition Object Size Algorithm Software V1.0, Tablet Deblistering Machine Control Software V1.0, and Tablet Packaging Information Verification Machine Intelligent Control Software V1.0. All copyrights were originally acquired with full rights.
JKBEAT's drug verification machines and automatic deblistering machines are classified as pharmacy automation equipment within the industrial automation equipment category. They do not fall under the medical device classification and do not require medical device registration certificates. This classification reflects their role as pharmaceutical logistics and quality control equipment rather than diagnostic or therapeutic devices.
Key technical parameters for the JKBEAT AI-Powered Drug Visual Verification Machine.
Note: JKBEAT's drug verification machines and automatic deblistering machines are classified as pharmacy automation equipment (industrial automation category), not medical devices. They do not require medical device registration. Specifications are based on standard configurations; actual parameters may vary depending on deployment requirements and drug database configuration. For detailed specifications tailored to your hospital's needs, please contact our team.
JKBEAT's tri-modal fusion technology is deployed and operational in Grade 3A hospitals across China. Whether you are evaluating drug verification solutions for the first time or looking to upgrade from barcode-only systems, our team can provide a technical consultation tailored to your pharmacy's workflow, volume, and compliance requirements.
| Accuracy | ≥99.99%, zero missed |
| Speed | 0.7 s/pack |
| Modalities | AI + Laser + Barcode |
| Drug Database | 50,000+ specs |
| Patent | ZL202410362593.X |