A novel pilot-assisted channel estimation model, Matrix-DenseNet, is introduced, which has a unique matrix-like structure consisting of five rows and six columns. Dense connectivity is incorporated within each row to enhance feature propagation and reduce parameter count. Additionally, deep sampling paths and feature feedback paths are set up across columns, creating a deep sampling feedback structure that further improves the extraction of multi-resolution features from the initial CSI tensor. Simulation results demonstrate that the proposed Matrix-DenseNet significantly improves the normalized mean square error (NMSE) and bit error rate (BER) performance of OFDM systems in high-speed environments.