<![CDATA[Waste is a universal problem. According to World Bank, there would be 3,4 billion tonnes of waste annually in 2050. To reduce waste in Indonesia landfills there are scavengers that sort the recyclable waste from the organic ones manually, but their capacity is limited. To increase the sorting rate there should be a sorting machine right from the source, which is civilian homes, or offices. To create such machine the writer will produce a sorting machine for organic waste and recyclable waste using the Resnet50 method. To train the Resnet50 model the writer use a dataset from Kaggle which consist of 22500 training and testing data. The Resnet50 model will be trained using 20 epochs with learning rate of 0,001 for the first 10 epoch and a learning rate of 0,0001 for the next 10 epochs, which resulted in a model with 99% accuration, 3% loss, 96% accuration validation, and 12% loss validation. The machine will then be tested with different object to camera lengths starting from 16 cm, 18 cm, 20 cm, 22 cm, 24 cm, and 26 cm. The best accuration is gained from the length of 20 cm and 22 cm with 85% accuration and overall average classification time of 1,17 seconds.]]>
Copyrights © 2022
