Cloud-side-end collaborative meat intelligent detection system and method

Abstract

The invention discloses a cloud-side-end collaborative meat intelligent detection system and a cloud-side-end collaborative meat intelligent detection method. The system realizes high precision, low delay and low resource consumption of meat detection through three-level cooperation of the mobile terminal, the edge device and the cloud server and a dynamic task scheduling strategy based on confidence, and is suitable for meat safety supervision in a large-scale and complex environment.

Description

Technical Field

[0001] This invention relates to the fields of intelligent detection and cloud computing, and in particular to a cloud-edge-device collaborative computing system and its working method for detecting the quality and authenticity of meat, aiming to achieve the best balance between detection efficiency, accuracy and cost. Background Technology

[0002] Existing technological shortcomings: Full cloud processing: High latency, strong network dependence, unsuitable for rapid on-site testing scenarios such as markets.

[0003] Full-terminal processing: Terminal devices have limited computing power, making it impossible to run complex models and resulting in low detection accuracy.

[0004] Rigid architecture: The task allocation strategy is fixed and cannot be dynamically adjusted according to network status, device load, and task complexity, resulting in low resource utilization efficiency.

[0005] Comparison documents: Patent CN108226408A: A meat quality testing device. This device uses traditional mechanical puncture and physicochemical testing methods (including mechanical components such as puncture needles, pipettes, and telescopic rods for moisture extraction and testing), without involving any cloud-edge-device collaborative computing or artificial intelligence algorithms. Its testing process relies entirely on fixed hardware devices completed locally, resulting in a cumbersome process, slow speed, inability to handle large-scale concurrent processing, and a lack of ability for intelligent task scheduling and dynamic resource allocation through software algorithms. Summary of the Invention

[0006] Technical issues: This addresses the problem that existing detection systems cannot achieve a dynamic optimal balance between latency, accuracy, and cost. Technical solution

[0007] This invention proposes a cloud-edge-device collaborative system with hierarchical decision-making and dynamic task scheduling.

[0008] System Architecture: 1. Mobile Terminal (End): Deploy a lightweight screening model. It is responsible for initial image acquisition and analysis. If the confidence level is higher than the threshold T1 (e.g., 98%), it returns the result directly; otherwise, it uploads the data.

[0009] 2. Edge Devices (Edge): Deploy standard precision models. Receive data uploaded by terminals and perform full-function analysis (authenticity + quality). If the confidence level is higher than the threshold T2 (e.g., 99.5%), return the result; otherwise, upload core feature data to the cloud.

[0010] 3. Cloud Platform (Cloud): Deploy ultra-high-precision models and gene databases. Conduct final reviews on difficult samples uploaded by edge devices, and can call gene libraries such as NCBI for comparison, send down the final judgment results, and use them for model updates.

[0011] Dynamic Scheduling Strategy: The scheduling center monitors network bandwidth, edge device load, and task queue length in real time, and dynamically adjusts thresholds T1, T2, and upload strategies. For example, when the network is congested, appropriately increase the T1 threshold to reduce the amount of uploaded data.

[0012] Technical Effects: ① Efficiency improvement: 95% of the samples are processed at the edge layer and below, and the average detection delay is reduced from 5 seconds in the pure cloud solution to 1.5 seconds.

[0013] ② Resource optimization: The cloud computing load is reduced by 70%, and the bandwidth occupancy is reduced by 80%.

[0014] ③ Accuracy guarantee: Through the ultimate review in the cloud, the overall accuracy of the system approaches 100%. Specific Implementation Modes

[0015] Example: Deployed in a certain farmers' market.

[0016] Terminal: Merchants use the mobile APP to take pictures of meat products, and the APP initially judges them as "low risk".

[0017] Edge: The market supervisor uses a hand-held instrument to conduct NIR spectral scanning on "low-risk" samples. The hand-held instrument judges them as "true, but the freshness is doubtful", and the confidence level 90% < T2, so the spectral characteristics are uploaded to the cloud.

[0018] Cloud: The cloud model conducts a review, confirms that the TVB-N value exceeds the standard, determines them as "spoiled meat", sends the results down to the hand-held instrument and the supervision platform, and triggers an alarm.

Claims

1. A cloud-edge-device collaborative intelligent meat detection system, characterized in that, include: A mobile terminal, configured to run the first lightweight model and perform initial screening; Edge computing devices are configured to run a second high-precision model and detect samples uploaded by mobile terminals; A cloud server is configured to run a third ultra-high precision model and to perform a final review of difficult samples uploaded by edge computing devices; The collaborative scheduling center is configured to dynamically allocate detection tasks based on network status, device load, and task complexity.

2. The system according to claim 1, characterized in that, The dynamic allocation strategy includes: mobile terminals and edge computing devices determine whether to upload data to the next level by comparing the confidence output of the local model with a preset threshold.