Dynamic Learning Rate Optimization for Waste Classification Using Convolutional Neural Networks and Custom Callbacks
S Jegadeesan, Rajashree Sridhar, S Sudharsun, M K Sujit, K. Venkata Sriram Subash · 2025
There has been increasing pressure for efficient Environmental management coupled with the rising problems of waste disposal, which has boosted research into automated waste classification systems. This paper presents the adaptation of Dynamic Learning Rate Optimization within a Convolutional Neural Networks (CNNs) design with Custom Callbacks to improve the categorization of wastes. Thus, the proposed dynamically adjusted learning rate based on validation loss allow the model work more efficient with variation of data distribution, and, therefore, reach higher accuracy on the images recognition as “Degradable” or “Non-Degradable”. We designed an Android application which takes pictures of the wastes and sends them to the backend server for classification using the CNN model. It reaffirms the practical time-sensitive disposal of waste through an interactive mobile application mode while enhancing the classification of waste to support the sustainable reuse of the environment.