River Trash Sorting System Using SSD-MobileNet V2

Indonesia experiences a buildup of trash on riverbanks due to the removal of trash from the riverbed to the riverbanks. After that, the trash is left for days without any follow-up. As a result, new problems arise an example environmental pollution. So there is a need for trash processing after the removal of riverbed trash. In this research, a deep learning-based trash sorting system using the SSD-MobileNet V2 model has been designed. This is a combination of MobileNet V2 and SSD models. This model is implemented on Raspberry Pi 4 model B. The FPS value obtained ranges from 1.46 to 2.40 FPS. Model performance testing is done by calculating the precision, recall, and F1-score values. The results show a precision value of 0.96 and a recall of 0.93. The F1-Score value is 0.94 which shows that the model has good precision and recall values. This model is also integrated with a sorting actuator in the form of a DC motor gearbox connected to a spur gear and rail gear. The results show that the system can sort trash that has a volume of more than 30 cm
³
with an accuracy of 80%. This model can be better developed by using a better microcomputer for greater FPS values. In addition, a better sorting actuator can be substituted. So that all trash of various sizes can be sorted by the system.