| Title |
An accurate and adaptable deep learning-based solution to floating litter cleaning up and its effectiveness on environmental recovery |
| ID_Doc |
25692 |
| Authors |
Li, QY; Wang, ZR; Li, GL; Zhou, CL; Chen, PY; Yang, CY |
| Title |
An accurate and adaptable deep learning-based solution to floating litter cleaning up and its effectiveness on environmental recovery |
| Year |
2023 |
| Published |
|
| Abstract |
Ending water pollution is urgent for environmental recovery globally. Floating litter cleaning up solution, as an essential strategy for reducing pollution, is formulated in a few studies. In autonomous waste collection systems like unmanned surface vehicles (USV), floating litter cleaning up is very challenging on natural water surfaces due to the complex environmental factors seriously degrading litter detection accuracy, lowering efficiency, or even causing the failure of collection. The impact of the cleaning-up solution on environmental recovery has yet to be discussed. To fill knowledge gaps, we propose an accurate and adaptable deep learning solution based on Faster RCNN to address this challenge through the first-time implementation of the attention mechanism at the C3 stage of ResNet50 to effectively extract the useful feature information and compress the negative influence of the complex environmental factors. We here first discuss the effectiveness of our cleaning-up solution in water pollution-reducing and waste recycling for resource-saving, to the best of our knowledge. The extensive exper-imental results on our self-built dataset show that our solution is superior to the state of the arts in the accuracy and adaptability of floating litter detection under the different complex scenes. Our solution improves the effectiveness of floating waste collection a lot. It significantly contributes to pollution mitigation, ecosystem, resource-saving, and human health. It is suggested that stakeholders should pay more attention to improving cleaning-up solutions. This work strives to shed light on it based on deep learning for environmental recovery and sustainability. |
| PDF |
|
