RESOLUTION-AWARE HETEROGENEOUS CPU–GPU COMPUTING FOR IMAGE PREPROCESSING

This study proposes a depth-context fuzzy neural network (DC-FNN) that explicitly combines local vertical context with a graded fuzzy representation before neural classification. The proposed DC-FNN achieved accuracy=0.7689, balanced accuracy=0.7680, macro-F1=0.7734, and weighted-F1=0.7691. Relative to the raw-feature MLP, accuracy increased by 14.02 percentage points; relative to a depth-context MLP without fuzzy expansion, the gain was 5.01 points. The adjacent-facies accuracy reached 0.9397, indicating that most residual errors occurred between sedimentologically neighboring facies. The results demonstrate that the principal benefit of the proposed formulation arises from coupling local depth context with continuous fuzzy partitioning rather than from increasing neural-network depth alone.