Overview of the iterative training and TL process used in SAD-CNN. The first training stage begins with 1764 images of edge-on galaxies at z = 0, where the network learns to classify images based on the presence of faint features. The second stage introduces 3664 additional images of edge-on galaxies at different redshifts, improving classification performance on galaxies observed at various cosmological distances. Finally, the third stage applies TL to a dataset of 5264 images of galaxies with different disc inclinations, refining the model’s ability to generalise across varying orientations. Throughout the process, feedback is incorporated at each step, allowing for progressive learning and improved classification accuracy.