Transfer learning is an important technology in machine learning. It is used to transfer knowledge from one domain to new problem task. It is used in many classification problems. Flower classification is one of the applications that uses transfer learning. Due to the large number of flower species that share similar shapes, appearances, or surrounding things like grass and leaves, flower categorization is a substantial task. We have proposed transfer learning model WMobileNet by enhancing traditional MobileNetV2 model. WMobileNet model uses AdamW and SparseCategoricalCrossEntropy algorithms as optimizers and losses respectively for flower classification on a Kaggle flower dataset. Because of adaptive momentum capabilities and learning rate Adam optimizer is popular. With L2 regularization AdamW is superior to Adam. Using gradient update process, it decouples weight decay. This results in increased convergence, better generalization, and consistent regularization. AdamW prevents interference with adaptive learning rates by separating the weight decay from the gradient calculation, which leads to more consistent and dependable optimization across various topologies. SparseCategoricalCrossEntropy is a loss algorithm that can be used for a number of multiclass classification problems. The classification accuracy is measured. It is found that WMobileNet performs better than CNN. This classification application can be utilized for identification of variety of patterns of flowers.