|
| Year | Paper | Advantage | Disadvantage |
|
| 2017 | [5] | (i) High speed of processing | (i) A lot of parameters required for training |
| (ii) High accuracy |
| (iii) Having an ability to intervene in big dataset images | (ii) Network architecture complex |
|
| 2018 | [8] | (i) Less processing times | The algorithm is dedicated for a small dataset like CIFAR-10; otherwise, the performance is not considerable |
| (ii) High accuracy with small image |
|
| 2019 | [6] | Highest accuracy for face image category | (i) Long chain of processing before classification |
| (ii) Lowest accuracy for classifying an image with a variant content like pizza category |
|
| 2021 | [9] | (i) Minimum number of epochs | (i) A million of parameters |
| (ii) High accuracy with image having small size | (ii) Weakness with dataset having large image |
|
| 2021 | [10] | Minimum time of training | Low accuracy for a dataset with many classes |
|
| 2021 | [41] | High accuracy | (i) Maximum number of parameters and epochs |
| (ii) High computation time |
|
| 2022 | [11] | (i) Maximum quantity of information in image signature | (i) Training time around 13 hours |
| (ii) Medium accuracy rate (top-1) | (ii) Accuracy improved observed only for top-5 accuracy measurement |
|
| — | [12, 19–21] | (i) Good accuracy | (i) Too much parameters |
| (ii) Minimum computation time |
|
| 2022 | Proposed method | (i) Minimum parameters required | (i) Number of epochs maximum is required |
| (ii) Minimum computation time (2.11 milliseconds for an image with small size (32 × 32)) |
| (iii) The architecture is always the same independently of image dataset | (ii) A bit difficulty to classify an image having important background |
|
