Progress in Deep Learning (DL), particularly Convolutional Neural Networks (CNNs), has significantly improved the accuracy of steganographic image detection. However, the applications of CNNs have several challenges, mainly due to insufficient dataset quality and quantity, the heightened imperceptibility of low payload capacities, and suboptimal feature learning procedures. This paper proposes an enhanced secret data detection approach with a CNN architecture that includes convolutional, depth-wise, separable, pooling, and spatial dropout layers. An improved fuzzy Prewitt approach is employed for pre-processing the images prior to being fed into the CNN. Experimental results show a significant outperformance over the state-of-the-art methods.