Figure 1.
Figure 2.
Figure 3.
Comparison to other methods
| Study | Year | Dataset | Method | Accuracy [%] |
|---|---|---|---|---|
| [6] | 2016 | TrashNet | SVM | 63 (test accuracy) |
| CNN | 22 | |||
| [7] | 2017 | TrashNet | Faster R-CNN | 68.3 (mAP) |
| [8] | 2018 | TrashNet | VGG-19 CNN | 88.4 (validation accuracy) |
| [9] | 2018 | TrashNet | Faster R-CNN based on Inception V2 | 84.2 (precision) 87.8 (recall) |
| Dense Net2 11 InceptionRes | 95 (test accuracy) | |||
| [10] | 2018 | TrashNet | NetV2 | 87 (test accuracy) |
| RecycleNet | 81 (test accuracy) | |||
| Pretrained VGG-16 CNN | 93 | |||
| [4] | 2018 | TrashNet | AlexNet CNN | 91 |
| KNN | 88 | |||
| SVM | 80 | |||
| [11] | 2019 | TrashNet | ResNet50 CNN with SVM Classifier | 87 |
| [5] | 2020 | TrashNet | MobileNet V2 | 98.7 |
| [12] | 2020 | TrashNet | MobileNet V2 | 97.6 (precision) 94.4 (recall) |
| Faster R-CNN based on Inception ResNet | 95.8 (precision) 94.4 (recall) | |||
| [13] | 2019 | LWW | Faster R-CNN | 86 (mAP) |
| [14] | 2018 | Custom dataset | Multilayer HybridCN N (MHS) | 98.2 (accuracy) 98.5 (precision) 99.3 (recall) |
| [15] | 2021 | TrashNet | Multilayer Hybrid CNN (MLHCNN) | 92.6 |
| [16] | 2021 | ISTWaste | Faster RCNN | 83 (test mAP) |
| [18] | 2021 | Wadaba | CNN | 74 (accuracy) |
| Own work | 2022 | TrashNet | Histogram | 94 (accuracy) |
The results of recognition using the Trashnet database
| No. | Type | FRR | FAR | Accuracy |
|---|---|---|---|---|
| 1 | Carton | 0 | 4 | 96 |
| 2 | Glass | 0 | 8 | 92 |
| 3 | Metal | 0 | 15 | 85 |
| 4 | Paper | 0 | 6 | 94 |
| 5 | Plastic | 0 | 2 | 98 |
| 6 | Trash | 0 | 1 | 99 |
| Average | 0 | 6 | 94 |
Results of experiment
| Range A | Range B | Accuracy [%] |
|---|---|---|
| 1-100 | 155-255 | 74 |
| 1-100 | 101-255 | 91 |
| 1-150 | 155-255 | 54 |
| 1-100 | 101-200 | 88 |
| 50-150 | 151-200 | 51 |
| 50-100 | 101-200 | 89 |
| 50-150 | 151-255 | 70 |
| 1-120 | 151-255 | 69 |
| 1-120 | 121-255 | 83 |
| 1-180 | 121-255 | 94 |