Fig. 1. Performance comparison between YOLOv10 model and other models^[18] (color online)

Fig. 2. Categories of training sample images(a) Ag, (b) Glass, (c) CeO₂, (d) Ag₂SiO₃, (e) TiO₂

Fig. 3. Annotated regions of interest (ROIs) in the training sample images(a) Ag, (b) Glass, (c) CeO₂, (d) Ag₂SiO₃, (e) TiO₂

Fig. 4. Metrics over 100 epochs (color online)

Fig. 5. Confusion matrix of the trained model test results

Fig. 6. Performance test results of Intel Xeon 8462Y+CPU

Fig. 7. Performance test results of NVIDIA A40 GPU

Fig. 8. Test of cropping and compression effects (a) Ag, (b) Glass, (c) CeO₂, (d) Ag₂SiO₃, (e) TiO₂

Fig. 9. Optimized deployment scheme in DAQ data stream

Fig. 10. DAQ framework deployment design

Fig. 11. Compression module design based on YOLOv10

Fig. 12. Testing of image results before and after compression on simulated data(a) Ag, (b) CeO₂, (c) Glass, (d) Single-crystal, (e) TiO₂, (f) Diamond

Fig. 13. Comparison of compression effects on simulated data

Fig. 14. Performance comparison of the compression module on different GPU platforms

Fig. 15. Design of multithreaded preprocessing and postprocessing deployment workflow

Fig. 16. Design of multithreaded deployment for DAQ framework

Fig. 17. Design of multi-mode compression for DAQ data processing