[ML] Image Recognition project with Big images

Hello people! We have many satellite images of a forest with different heights of vegetation. From them we have extracted 20 big square areas called AOIs (areas of interest) in tiff format (after we have cleaned them from inappropriate data), 8 gb each, with labels and everything. We need to classify those squares in 5 classes, according to their height, with 3 methods: A) With pre-trained networks: 1 - LZW compression 2-NN-1m compression B) Without pre-trained networks: Instead of writing a network from scratch and scrutinizing the ideal depth of the network, number of parameters within each layer and other hyperparametrs, you can simply find the code of a pre-trained network on keras and copy paste it with some modifications. We have implemented 1 method only for 8 of the 20 AOIs. We want them all!! You will either need a great CPU, a great GPU, or a smart method like we did, to train it. We will also need full documentation of the implementation, accompanied with screenshots. I will send the full description in chat and I will be happy to answer any questions. Best