Article Preview
TopIntroduction
According to the World Bank and the World Economic Forum, the world is producing more than two billion tons of solid waste per year, with almost 175 million tons in plastic waste. Much of this waste is either dumped in oceans or in landfills, thus polluting the environment or endangering marine life. These numbers are expected to increase by 70% annually within the next 30 years, due to economic development, the rapid growth of populations, and rapid urbanization, if recycling and waste management stays at the current rate (Dias et al., 2019). For example, the Middle East and North Africa produced 129 million tons of waste in 2016 (World Bank, 2021), and they are expected to produce 255 million tons in 2030. The largest increase is in Sub-Saharan Africa, in which waste production will increase from 174 million tons in 2016 to above 516 million tons in 2030 (Jin et al., 2017; Diffenbaugh & Burke, 2019).
Waste recycling is of great importance, not only for obvious environmental reasons but also for sustainable economic growth, mainly in modern societies. Moreover, waste management plays a great role in reducing the overall effects of both solid and non-solid waste. Many countries have adopted new and emerging technologies to help in sorting and classifying the different types of solid waste to make recycling more efficient. For example, Shanghai deployed more than 2,000 artificial intelligence (AI) trash bins for waste sorting for the sake of sustainable development, and many Chinese cities are following in the footsteps of Shanghai. Eventually, many developing nations can benefit immensely from such technologies (Lu & Sidortsov, 2019).
AI and artificial neural networks (ANNs) are at the center of these technologies for automating the process of classifying the different types of waste. Both disciplines are motivating researchers, including this research, to employ such techniques to achieve automation processes for classifying and separating waste for better recycling. Identifying and measuring the trash recyclability level is important for the designing process, policy making, and environmental protection. However, existing methods for automating the measurement of recyclability are not accurate, because they are subject to many factors such as material composition, background, size, and shape (Chancerel & Rotter, 2009).
This research aims to develop a robust hybrid deep learning model that uses CNN and FIS that would be developed by using human experts. Adding adjustable fuzzy rules will increase the accuracy and will be dynamic to suit the different waste types. The results will be compared with three popular CNN architectures AlexNet, VGG-16, and ResNet50.
The contribution of this research can be summarized in the following points:
- •
Creating a new hybrid learning model, consisting of several layered of CNN and FIS system:
for end-to-end classification using a deep learning model, a large amount of data is required, which is often not always available, so we resort to using transfer learning to generate the initial numeric scores using standard CNN layers, then rules can be developed based on these numeric scores to produce both efficient and accurate classification model.
- •
Applying several experiments with the proposed model on waste public dataset (Sekar, 2019):
the results of the proposed model were compared with other popular CNN architectures AlexNet, VGG-16, and ResNet50 models, the obtained results outperform the existing models by a significant margin.
- •
Using fuzzy inference system for Measuring Waste Recyclability:
FIS has the advantage of using adaptive priority based on the numeric scores as well as human experts in the field for each sample to be predicted, and hence performs better than traditional CNN models, both rules and fuzzy membership functions can be adaptive according to the target domain.