................

Deep neural networks (DNNs) are extensively utilized in hyperspectral image (HSI) classification tasks due to their higher and more comprehensive learning ability through their layer structure. The optimizer plays a crucial role in the convergence of a deep neural network. Building upon the foundational Fast Dense Spectral-Spatial Convolution (FDSSC) network, which efficiently extracts spectral and spatial features for classification, this study investigates a crucial yet often overlooked aspect: the choice of optimizer in deep neural networks. This paper examines the role of the optimizers in the convergence of such networks for hyperspectral image classification. While the original FDSSC framework employed RMSProp to achieve high classification accuracy, we extend this work by systematically analyzing the impact of multiple optimizers—SGD, ADAGRAD, ADADELTA, RMSProp, and Adam—on model convergence and performance. Through comprehensive experiments on benchmark datasets such as Kennedy Space Center, Pavia University and Indian Pines, we show that the Adam optimizer significantly enhances overall accuracy and accelerates convergence compared to its counterparts. Our customized FDSSC (cFDSSC) model, integrated with Adam, consistently outperforms the original setup and provides practical guidance for optimizing deep learning models in HSI classification. Furthermore, the number of epochs needed to achieve the best performance for the optimizers is also investigated. Optimizer performance was assessed over varying epochs, revealing that optimal accuracy and convergence speed depend on the choice of optimizer.