Event cameras possess advantages such as ultra-high dynamic range and ultra-low latency. Extracting effective spatio-temporal features from the output data of event cameras through event stream segmentation, filtering, and event representation is crucial to leveraging these advantages. While existing event representation methods based on timestamps and exponential kernel functions for calculating time surfaces can preserve more informative details in events, they still face issues like high event redundancy and susceptibility to noise events. Addressing the high redundancy in existing event stream segmentation and filtering methods, this paper proposes a novel event downscaling algorithm based on density sorting. This algorithm analyzes the spatio-temporal neighborhood relationships within the event stream to calculate spatio-temporal correlation density and performs density sorting accordingly, thereby reducing redundant events and minimizing the consumption of computational resources. Furthermore, to address the vulnerability of existing event representations to noise events, this paper introduces an event data representation method based on spatio-temporal neighborhood correlation for denoising the time surface. This method considers spatio-temporal correlations to form event clusters on the time surface, effectively filtering out valid events and enhancing the signal-to-noise ratio while reducing computational complexity. The proposed methods have achieved state-of-the-art (SOTA) classification accuracy on three mainstream neuromorphic datasets. In summary, this paper focuses on the research of event stream data dimensionality reduction and event representation for event camera object classification, effectively improving the efficiency and accuracy of event camera object classification.