SPRADC9 july   2023 AM62A3 , AM62A7

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1Introduction
    1. 1.1 Defect Detection Demo Summary
    2. 1.2 AM62A Processor
    3. 1.3 Defect Detection Systems
    4. 1.4 Conventional Machine Vision vs Deep Learning
  5. 2Data Set Preparation
    1. 2.1 Test Samples
    2. 2.2 Data Collection
    3. 2.3 Data Annotation
    4. 2.4 Data Augmentation
  6. 3Model Selection and Training
    1. 3.1 Model Selection
    2. 3.2 Model Training and Compilation
  7. 4Application Development
    1. 4.1 System Flow
    2. 4.2 Object Tracker
    3. 4.3 Dashboard and Bounding Boxes Drawing
    4. 4.4 Physical Demo Setup
  8. 5Performance Analysis
    1. 5.1 System Accuracy
    2. 5.2 Frame Rate
    3. 5.3 Cores Utilization
    4. 5.4 Power Consumption
  9. 6Summary
  10. 7References

5.1 System Accuracy

Several experiments were implemented to extensively test the accuracy of the entire system which includes the accuracy of both the yolox-nano-lite model, trained on the defect detection data, the object tracker, and the graphical dashboard. The experiments test the defect detection application live on testing samples (ring terminals). A collection of samples with predefined combination of classes (good, half ring, no plastic, and no ring) are placed the rotating table to simulate moving on a conveyor belt. The application is used to detect the samples and the results shown on the dashboard are compared with the actual statistics of the samples.

Table 5-1 shows the details of one experiment which includes a total of 50 samples with 20 % defected samples distributed as following: 3 half ring, 5 no plastic, and 2 no ring.

Table 5-1 Accuracy Experiment Details of Defect Detection Application (it shows 100 % accuracy with 50 samples and 10 repeated tests)
Class Ground Truth
No. and [%]		 Application Results After 10 Rounds Accuracy
Total Samples (Not a class) 50 500 NA
Good 40 [80%] 400 [80%] 100 %
Half Ring 3 [6%] 30 [6%] 100 %
No Plastic 5 [10%] 50 [10%] 100 %
No Ring 2 [ 4%] 20 [4%] 100 %

The samples are randomly placed on the rotating table and the application is used to detect them. The table with the samples were rotated for 10 rounds for repeatability assurance. Figure 5-1 shows the dashboard at the end of the tenth repeated tests. Comparing the results on the dashboard generated by the application with the ground truth input samples showed that the application successfully detected all good and defected units in all ten rounds as shown in Table 5-1.

GUID-20230630-SS0I-KL9D-MJML-46JW96VJDXLB-low.png Figure 5-1 Dashboard Results of the Accuracy Experiment With 50 Samples and 10 Repeated Tests (the defect detection application achieved 100 % accuracy)