Intelligent solid waste classification using deep convolutional neural networks

Parallel to the rapid growth of the population, the rate of consumption is increasing all over the world; this causes significant increases in the amount of waste. Thanks to the recycling of waste, not only environmental pollution is prevented but also a great contribution to the economy is made. One of the basic conditions for ensuring maximum performance in recycling processes is the classification of wastes according to their contents. At this stage, minimizing the human factor is an important issue in terms of time, labor, and performance of recycling facilities. In this research, paper, glass, plastic, and organic waste pictures obtained from the external environment were classified with the help of machine learning techniques. In classification, four- and five-layer deep convolutional neural networks algorithms were used. According to the results of the research, five-layer architecture was able to distinguish the wastes with a 70% accuracy rate. In the research, as the number of layers decreased, the performance values of the networks decreased. In the four-layer architecture, wastes could be separated by a rate of 61.67%. In both network architectures, the accuracy rate in differentiating plastic wastes from other wastes was found to be lower. The accuracy rate in the classification of plastic wastes was determined as 37% and 56.7% in four-layer and five-layer DCNN architectures, respectively. In the research, organic wastes were distinguished with higher accuracy compared to other wastes. The accuracy rate in the classification of organic wastes was determined as 83% and 76.7% in four- and five-layered DCNN architectures, respectively.

Automated summary

In this research, machine learning techniques were used to classify paper, glass, plastic, and organic waste images obtained from the external environment. The five-layer architecture was able to distinguish wastes with a 70% accuracy rate, while the four-layer architecture achieved a 61.67% accuracy rate. The accuracy rate for differentiating plastic wastes from other wastes was found to be lower, at 37% and 56.7% in four- and five-layer DCNN architectures, respectively. Organic wastes were distinguished with higher accuracy, at 83% and 76.7% in four- and five-layered DCNN architectures, respectively.