ABSTRACT
Image segmentation has been a significant phase of digital image processing. Being a critical phase, it is necessary to ensure accuracy in object detection is achieved at initial stage for maximum efficiency unlike post processing optimization. On contrary to traditional image processing, learning based image processing has been very effective due to recent advancements delivered by deep learning based neural network models. Optimization in neural network architecture has always been an intrinsic part that can address computational cost especially due to their vast learning time. In this research optimization of learning time is challenged with the association of Adam optimizer specific to effectively balancing the momentum with learning rate. The paper uses image dataset containing car number plates captured in multiple traffic stops for validation of optimizer using YOLO-V3 architecture approach for quantification of success. Evaluation of the proposed experiment is measured based on the mAP score and the convergence time while training the model.
