Fig. 1. GRNet model general architecture
Fig. 2. Change diagram of real frame of infrared the gas leakage
Fig. 3. Visualization of gas leakage candidate box clustering analysis
Fig. 4. Aspect ratio visualization results of candidate frames for gas leak infrared detection dataset
Fig. 5. Mosaic data enhancement
Fig. 6. Gamma transform rendering
Fig. 7. Image enhancement preprocessing schematic
Fig. 8. Fitting diagram of real box and prediction box
Fig. 9. CIoU positioning loss
Fig. 10. Structure of feature extraction network reconstructed by RepVGG module
Fig. 11. RepVGG module structure diagram
Fig. 12. The final test result of the GRNet network model
Fig. 13. Visualisation of ammonia leak concentration distribution
Fig. 14. Visualisation of the overall interface of the infrared gas leak detection system
Detection layer | Before clustering | After clustering |
---|
Detection layer 1 | (10,13),(16,30),(33,23) | (11,10),(29,12),(34,29) | Detection layer 2 | (30,61),(62,45),(59,119) | (52,61),(62,18),(64,38) | Detection layer 3 | (116,90),(156,198),(373,326) | (91,38),(115,22),(201,45) |
|
Table 1. The size of the initial anchor frame of the three detection layers before and after clustering
Hyperparameter name | Hyperparameter values |
---|
Batch size | 16 | Learn rate | 0.01 | Epoch | 400 | Momentum | 0.937 | Policy | Cosine annealing strategy | Weight decay | 0.000 5 |
|
Table 2. Hyperparameter configuration
Model | Params/MB | Model size/MB | Time/ms | mAP/% |
---|
YOLOv5s-GIoU | 7.05 | 14.40 | 3.80 | 92.20 | YOLOv5s-CIoU | 7.05 | 14.40 | 3.60 | 92.80 |
|
Table 3. Comparison of the verification performance of different localization loss functions for leaky targets
Model | Params/MB | Model size/MB | Time/ms | mAP/% |
---|
YOLOv5s-GIoU | 7.05 | 14.40 | 3.80 | 92.20 | YOLOv5s-CIoU | 7.05 | 14.40 | 3.60 | 92.80 | YOLOv5s-CIoU-MG | 7.05 | 14.40 | 3.60 | 93.40 |
|
Table 4. Comparison of the network performance before and after image pre-processing
Model | Params/MB | Modelsize/MB | mAP/% |
---|
YOLOv5s-GIoU | 7.05 | 14.40 | 92.20 | YOLOv5s-GIoU-Km | 7.05 | 14.40 | 93.20 | YOLOv5s-CIoU-Km | 7.05 | 14.40 | 93.70 | YOLOv5s-CIoU-MG-Km | 7.05 | 14.40 | 94.00 |
|
Table 5. Comparison of network performance before and after clustering
Model | Params/MB | Model size/MB | Time/ms | mAP/% |
---|
Conv-YOLOv5s | 7.05 | 14.40 | 3.80 | 92.20 | DWConv-YOLOv5s | 4.08 | 8.50 | 3.10 | 93.90 | ODConv-YOLOv5s | 5.50 | 11.30 | 3.70 | 94.20 | GRNet | 5.47 | 11.30 | 3.40 | 94.90 |
|
Table 6. Different modules embedded network performance verification comparison
Model | GFLOPs | Params/MB | Model size/MB | Time/ms | mAP/% |
---|
YOLOv3 | 154.7 | 61.50 | 123.40 | 11.70 | 92.70 | YOLOv3-tiny | 12.90 | 8.70 | 17.40 | 3.40 | 37.40 | YOLOv5s | 16.30 | 7.05 | 14.40 | 3.80 | 92.20 | YOLOx | 26.64 | 8.94 | 71.90 | 8.43 | 89.78 | GRNet | 15.60 | 5.47 | 11.30 | 3.40 | 94.90 |
|
Table 7. Comparison of the validation accuracy among various network models
Model | Speed/(frame·s-1) | mAP/% |
---|
YOLOv3 | 0.76 | 92.70 | YOLOv5s | 2.43 | 92.20 | GRNet | 3.03 | 94.90 |
|
Table 8. Embedded platform deployment test results