<p>Satellite imagery used for defense and disaster management contains crucial information which subjected to unauthorized access can lead to data tampering or theft. The conventional encryption techniques offer security but distort the image quality making the encrypted data useless for classification. This raises the need for a privacy-preserving classification framework that integrates encryption and deep learning methodologies. To preserve the secure images, this work proposes a Hybrid Block Scrambling Encryption (BSE) and Somewhat Homomorphic Encryption (SHE) (HBS²E technique). The main motivation of this work is to devise a novel hybrid encryption-based optimized deep learning approach to prevent the sensitive space image from potential breaches, increase classification accuracy, and balance encryption robustness. Meanwhile, the encrypted images are forwarded to the image classification phase known as Dynamic Hippopotamus (DH) based Modified Single Shot Multibox Detector (MSSMD) which sophisticatedly detects and classifies the images for the respective applications. The dynamic levy flight model is integrated with the hippopotamus algorithm to form the DH algorithm to solve the issues associated with satellite image classification. The proposed model achieves encryption/decryption accuracy of 98% on the EuroSat dataset and reduces the processing time by up to 15% when compared to the existing techniques. The obtained outcomes for the proposed work are encryption accuracy of 98%, Decryption accuracy of 97%, classification accuracy of 97%, and security level of 98%. While comparing the outcomes the proposed technique can secure the sensitive data of the satellite images more than the previous works.</p>