Multi-spectral satellite imagery (MSSI) offers a crucial means of collecting environmental and geological data from dangerous, remote locations. However, the substantial volume of such imagery complicates its storage and transmission, making efficient image compression indispensable. Traditional compression methods such as Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT), Consultative Committee for Space Data Systems (CCSDS), Bandelet, and JPEG 2000 have provided high reliability, simple implementation, and effectiveness, yet they lack sufficient adaptability. In contrast, Machine Learning (ML) offers a more adaptable, albeit complex and data-intensive, alternative. This paper proposed a hybrid approach, called AMIC, that combines the robustness of traditional methods with the flexibility of ML, using DWT to quantize image bands and ML to refine this process, thereby enhancing the compression quality with minimal computational load. The AMIC performance demonstrates high fidelity in compressed images, with Peak Signal-to-Noise Ratio (PSNR) values ranging from 30dB to 42dB and Structural Similarity Index (SSIM) values above 0.78, ensuring excellent preservation of image details. The approach significantly optimizes compression and decompression times, achieving a reduction to 43.26% of the original size. This hybrid method ensures high-quality satellite image compression for improved transmission and storage. It broadly impacts real-time data processing across various fields, including environmental monitoring and remote sensing research.