Abstract:
The current paper deals with management of urban food solid waste (FSW), which is the high-priority issue of health, environmental sustainability, and activities of cities. It is the nature of the issue brought into the limelight at sustainability courses that brings in questions about the inability to have a successful resolution. The problem is relatively simple, but at the same time it gets complicated due to the issue of scaling, contamination, and classification, which Noske and Funke display in their article on the sorting of glass beadsin terms of color (Figure 009). In our laboratory experiment, we were able to see that manual sorting is rather tedious and time-consuming and energy-intensive, and this fact supports the necessity of automating the process. The paper willsuggest a deep-learning-based solution to the problem of automatic labeling of food waste based on a large amount of high-resolution photographstaken in various urban conditions. The research approach is a novel and practical method ology with substantiated data. DenseNet201 was modified through transfer learning, large data augmentation, and balancing methods to remain general to overcome skewed data distributions. The dataset consists of more than 600 annotated images of five categories, which include buffet food waste, generalfood waste, restaurant food waste, vegetable waste, and wedding food waste that captures real-life conditions of waste in urban areas. After optimization, DenseNet201 was compared to InceptionV3, EfficientNetB2 and MobileNetV2, but trained with the same settings. DenseNet201 was found to be more precise and recall as well as achieve better validation accuracy with over 86 percent, being more accurate and reliable than the competing architectures, especially underrepresented classes. The suggested system has potential in application in smarter urban waste-management projects, university sustainability projects, and resource-optimization endeavors, which makes it a strong contender in a computer science capstone project.
