|
| S. No | Ref | Model used | Merits | Demerits |
|
| 1 | [1] | CNN Model | This model detects the damage location and damage severity with a validation accuracy of 89.06% | (i) But the model is implemented with the static condition |
| (ii) The bridge is temporarily closed and determined using static loading conditions |
| (iii) This can be expanded for the dynamic loading condition |
|
| 2 | [2] | Fast R-CNN | Data augmentation gives a better mean F1-Score of 6.255 when compared with R-CNN 6.174 | Much weightage is given for the preprocessing techniques model evaluation training is not discussed |
|
| 3 | [14] | Neural network-based damage detection method | Efficient method for pattern detection | (i) A considerable amount of false alarms |
| (ii) Structural damages of the bridges are not identified properly |
|
| 4 | [19] | FCN-based road surface anomaly detection model | The road anomalies and the location detection can be given approximately | But the depth of the anomalies cannot be determined by the proposed model |
|
| 5 | [27] | A black box camera is used to detect a pothole | (i) Potholes detection | (i) Black box camera uses a limited computing environment. The intensity of the light affects the false detection |
| (ii) Suitable for pavement management system |
|
| 6 | [28] | Decision Tree heuristic algorithm | Pavement crack detection | (i) More processing time due to preprocessing of images |
| (ii) Horizontal and vertical classification of pictures needed for further crack detection |
|
| 7 | [29] | SSD inception V2 and SSD MobileNet | Developed for classifying eight types of road damages | (i) Could achieve a precision of 77% and recall of 71% only |
| (ii) Dataset consists of comparatively fewer pothole images |
|
| 8 | [30] | Vibration sensor-based method | It can be implemented using smartphones with sensors | (i) The results are with low accuracy |
| (ii) The vehicle’s suspension and speed affect the amplitude vibration and the accuracy |
|
| 9 | [24] | Machine learning (ML)-based approach | Less computational and time complexity compared to deep learning models | (i) Hand-picked feature extraction using traditional image processing methods |
| (ii) Machine learning models produce less accurate results than deep learning models |
|
| 10 | [20] | Image processing-based crack detection | Cost-efficient and fast | (i) There is only one form of damage found |
| (ii) This leads to a high error rate in the presence of poor illumination and distortion |
|
| 11 | [25] | Machine learning (ML)-based approach | Less computational and time complexity compared to deep learning models | (i) Only for pothole detection |
| (ii) The results of machine learning models are comparatively lower than the deep learning models |
|
| 12 | [26] | Encoder decoder architecture of SegNet DeepCrack | This can be used for edge detection and images and vessel detection in retina images | The accuracy can be improved; it produces a 0.87 F-measure |
|
