Pretrained nets with 3+ channels

[stephandooper]

keras pretrained netwroks are incompatiable with more than 3 channels, find a potential fix for this?

[stephandooper]

A fix can be made in two ways:

• Extend the pre-made networks to 5 channels and do not include pre trained weights (see https://stackoverflow.com/questions/49546922/keras-replacing-input-layer)

• Adding a conv2d and connecting it to a pretrained network, see https://stackoverflow.com/questions/55944771/change-number-of-input-channels-to-pretrained-keras-applications-model