Steganography is the concept of embedding or hiding secret information into a cover image by maintaining the visual quality. Various algorithms are designed to classify stego images but the race still continues between Steganographer and Steganalyser. Advances in deep learning provided a solution to detect stego images. In this article, we coin a new paradigm to detect stego image as a three-step process with the following repercussions: (1) employing preprocessing step to enhance the input image, (2 feature extraction using the Mustard honey bee optimization algorithm and, thus, the extracted features will be dimensionally reduced (3) by classification using HSVGG-based CNN. Experimentation carried out on ALASKA2 data set and the results were compared.