Transfer learning is a widely utilized and crucial technology that facilitates the transfer of knowledge from one domain to solve new problems. It is extensively applied in various classification tasks, including flower classification, which presents unique challenges due to the extensive variety of flower species, various of those have similar shapes, appearances, or share environmental features namely, the leaves and grass of the different flowers. In this paper, we introduce NMobileNet, an enhanced version of the traditional MobileNetV2 model, specifically optimized for flower classification. NMobileNet utilizes the Nadam optimizer and SparseCategoricalCrossEntropy loss function to achieve superior performance on the Kaggle flower dataset. The Nadam optimizer is an extension of Adam optimizer is utilized in the paper. It includes adaptive learning rates with Nesterov momentum, offering advantages such as faster convergence, improved performance on complex non-convex cost surfaces, and smoother optimization paths. Sparse Categorical Cross Entropy is also an efficient loss for any multiclass classification problem. We demonstrated classifying performances of the NMobileNet and compared with classical CNN design. The experiment outcome showcase that the proposed NMobileNet has superior performance than CNN in this task, which illustrates the efficiency in recognizing diverse flower patterns. This is a promising alternative in areas such as botany, agriculture, and other situations where flower identification is essential.