This study introduces an advanced encryption technique for multi-spectral images that combines chaotic systems and a gravitational model to enhance security. The method tackles challenges like high dimensionality and inter-band correlations through a multi-layered approach. By using bit-plane decomposition, it achieves precise data manipulation, while a hybrid chaotic system combining 2D Logistic-Tent-Modulated Map and 1D Sine-Cosine-Sawtooth Map ensures high-quality randomness for pixel and spectral band scrambling. Additionally, a gravitational model-based diffusion process dynamically modifies pixel intensities, further strengthening encryption, dynamic image-dependent key generation ensures unique encryption keys for every image, enhancing resistance to brute-force attacks. The decryption process is fully reversible, ensuring accurate image reconstruction. Experimental results highlight the method's high sensitivity to initial conditions, strong defence against statistical and differential attacks, and efficient handling of multi-spectral data. This makes it a secure and scalable solution for applications like remote sensing, medical imaging, and defence.