This paper addresses the problem of blind digital modulation identification in time-selective Multiple-Input Multiple-Output channels. Our objective is to recognize modulation schemes in highly-mobile communication environments, for military or high-speed railway applications, without signal knowledge or Channel State Information at the receiver. The proposed identification process is based on Blind Source Separation (BSS) and feature classification. We introduce a sliding window technique for the BSS of a faded-mixture to overcome the effect of the high mobility. Then, to improve the recognition of modulation schemes, we adopt a specific multi Artificial- Neural-Network (ANN) classifier, where each ANN is trained to be used within a particular Signal-to-Noise Ratio range. The proposed identifier has a good probability for achieving correct identifications under high velocity for typical carrier frequency and bandwidth.