Wavelet-enhanced PSF-guided morphological convolution. After decomposing the input into multi-scale frequency components via wavelet transform, the model applies a cross-shaped convolutional structure that conforms to the morphology of astronomical PSFs, particularly the extended arms. This design aggregates dispersed information along the PSF pattern toward the central region, thereby enhancing weak, extended features. The receptive field visualization illustrates how this morphological structure enlarges the perception range and improves sensitivity to cross-shaped sources.