WasteInNet: Deep Learning Model for Real-time Identification of Various Types of Waste

The global challenge of waste management is becoming increasingly pressing due to population growth, urbanization, and industrialization. Detecting and classifying different types of waste materials is essential for efficient and sustainable waste management practices. This research aims to create a deep learning model for real-time waste detection that is categorized by type and emphasizes the importance of accurate waste identification. Various waste detection techniques have emerged, including visual, chemical, and technological methods. Visual inspection remains the fundamental approach, relying on human operators to sort waste based on appearance. However, limited human perception and increasing waste volumes require more automated solutions. Computer vision, which utilizes machine learning algorithms, has become well-known for its ability to classify waste based on visual attributes. This technology can differentiate between recyclable, non-recyclable, hazardous, and organic waste, thus providing a more efficient and accurate alternative to manual sorting. The research method starts with data collection, preparation, modeling, and evaluation. The research results are based on the overall performance of the test dataset, achieving a precision of 0.801, mAP@0.5 of 0.868, and mAP@0.5:0.95 of 0.618. The refined model results showed higher detection efficiency across several target categories, with the paper category showing the highest average precision (AP) value at 97 %. The model's average precision (mAP) was determined to be 86.8 %. The model that has been created can identify types of waste well. Despite the high performance, the results obtained from the test data set still require further improvement to overcome the challenges that hinder the accurate detection of various types of waste.